You are viewing [info]bjs_redhat's journal

About this Journal
I'm a long-time IT Consultant and former (of no glory to speak of) engineer, support engineer and/or pseudo-IT guy masquerading as an engineer (or some might say an engineer masquerading as a technologist). This is my blog. I do not speak for my employer(s), client(s), past employer(s), future employer(s) or other employer(s) that I may or may not be aware of. This is my blog. Read it. Enjoy it. Print it out. Don't finish half of it. Kill lots of trees. Most of my posts are usually long, winded and you probably dump them about half-way through. But every now and then all the information will be useful to at least a handful of people. ;)

Also see my LinkedIn Profile (bjsmith).

Current Month
 12345
6789101112
13141516171819
20212223242526
2728293031
May. 21st, 2012 @ 01:08 pm See No Downstream, Hear No Upstream
Current Location: United States, West Virginia, Fairmont
Current Mood: amusedamused

I've bit my tongue.  I've largely ignored many things.  But at this point, I'm seeing a pattern.  Several commercial Linux vendors continue to market against Red Hat Enterprise Linux.  That's to be expected.  But the pattern that's becoming clear is how they are not marketing against Red Hat itself, just Red Hat Enterprise Linux.

Bayer commercial gone even worse

Reminds me of a variation of an old Bayer commercial.  You know, the one on the island where they ask you to pick between Bayer, Tylenol, Extra Strength Tylenol and Advil?  And the result is Bayer #1, but anyone who attended grade school knows that if you add Tylenol and Extra Strength Tylenol, Bayer loses.  Well, it's kinda like that.

I call this "See No Downstream, Hear No Upstream" marketing.  Unlike Bayer, who openly disclosed they split Tylenol from Extra Strength Tylenol, this is now like breaking out Baby Tylenol (like 'Downstream' for arguments sake) and Extra Strength Tylenol (like 'Upstream' for argument's sake) from regular Tylenol on the survey.  But when the results come out, they don't tell anyone about the first two, only Tylenol is listed.  Even if it doesn't add up to 100%, one cannot see (and add in) the other two, because they are not disclosed.

Even more humorous (to me at least) is that I'm replacing 'Evil' with the words 'Downstream' and 'Upstream' from Red Hat.  I don't know anyone who considers those in the Downstream and Upstream to be 'Evil,' much less Red Hat's efforts towards them, so I get an added chuckle in the phrasing.  ;)

See no downstream:  EL Rebuilds

It is occurring regularly with the leader of one commercial Linux company, who has a distribution that is unpaid and offers a paid support model.  The leader and commercial company will go unnamed largely because I do not believe it is healthy to focus on one individual who honestly wants to help the community overall.  He, and the organization he founded, honestly want to help Linux grow, and I will always recommend everyone help fund this organization if you believe in the distribution they release.  I also have several colleagues who work for this company as well.

But the "See No Downstream" issue came about when he stated his distribution, with an unpaid option, now has more marketshare than the paid Red Hat Enterprise Linux in the cut-throat, often unpaid Linux websphere.  He's correct, of course, in this one market.  No one took issue with that.  No one debated that point.

Of course, he completely ignored the fact that "Enterprise Linux (EL) Rebuilds" (e.g., CentOS, Scientific Linux, etc...) are downstream from Red Hat Enterprise Linux, and have an even bigger share.  The "EL Rebuild" maintainers and users didn't like this omission, and let people know about that.  Of course, this then caused a counter-argument that his distribution is actually downstream from Debian.  Debian is a long-standing, democratic-organized Linux development effort (one where even I was a maintainer, of no real note, for a few years myself), which has the largest marketshare, more than all of the "EL Rebuilds" combined.

Now this is where the marketing, counter-marketing and counter-counter-marketing gets really subjective.  How much does this one distribution contribute upstream to Debian compared to how much "EL Rebuilds" gain from being downstream from Red Hat Enterprise Linux?  I know where some Debian people stand and I know where some "EL Rebuild" people stand.  But it's not my place to evaluate such and, frankly, I really don't like going here at all.  One has to step back and recognize the original issue with the marketing ... "See No Downstream" from Red Hat.  It's best just to leave it alone.

Of course, even that consideration doesn't even start to talk about Red Hat's upstream work, including the Fedora Project.  Red Hat employees contribute upstream and to Fedora as a paid job function.  But instead of diving into this case more on that, there's an even better one ...

See no Upstream:  Fedora is stable, but not Enterprise Linux

Recently a newer, commercial Linux vendor has been throwing a lot of marketing words against Red Hat.  In this case, I won't dance around who it is ... Oracle.  Now several Oracle consultants know me ... and well ... let's just say I get to smile a lot.  But aside from any points of great humor, there's a lot of non-sense that goes on that I just ... well ... I think this example will put it to rest.

Oracle recently blogged regarding its Unbreakable Enterprise Kernel (UEK).  It's beyond a perfect example of "See No Upstream," but we'll get to that in a moment.  It does quite a bit to undermine their entire positioning.  Some may not see it, but I believe most who are familiar with the value of Enterprise Linux will once I go through many of the statements.

Let's start with this paragraph.

'A CTO of another company a little while back implied on twitter that UEK is too current and reduces reliability because the code is new. Can I point something out here? One thing we decided to do, with UEK, was to not backport features but instead just focus on stability. The goal is to stay close (but not too close as we don't ship a development tree) to the "Linus kernel" and focus on testing the heck out of it and fixing any bug that we would find, any regression.' 

This sounds like a worthwhile effort.  Oracle offers a kernel just behind leading edge development, trying to balance features against stability.  Oracle spends the time testing and ends up fixing bugs and regressions.  In fact, they state the value of this for the community to finish off the paragraph.

'This is good for the Linux community because when we do find a problem, we fix it and it's immediately also relevant to the mainline kernel. This helps us and the kernel community. We spend time testing and fixing, we develop on mainline and we don't waste time backporting new stuff into old versions.'

Indeed.  The strength of Linux is the working with the community.  As Oracle puts it well at the very end ...

'It's a symbiotic relationship'

No argument.  No argument at all.  Oracle is very much correct.  But this is the "See No Upstream" marketing, as I will now break down.

Oracle is putting a people on the same thing that Red Hat also puts even far more people on.  But the difference here is that Oracle releases such an effort as "Enterprise" -- namely Unbreakable Enterprise Kernel (UEK) -- while Red Hat has a different name for it.

Anyone?  Anyone?  Red Hat's sponsored Fedora(TM) Project.

Fedora is stable, right down to the massive efforts into the kernel.  Fedora is unit and integration tested.  Fedora even backports some fixes, and avoids immediately rebasing with the upstream.  I have worked at many enterprises who also use Fedora.  Fedora is popular in hosting and in other areas as well.

Red Hat employees on the kernel and other, core subsystems.  They are tracking and even patching upstream kernel.org, and backporting to recent kernel releases in Fedora, as well as even older releases in Red Hat Enterprise Linux.  It's also a 'symbotic relationship,' and to be subjective, far more fruitful with many.  But that aside, what is my point here?

Ever use a Fedora kernel under Red Hat Enterprise Linux?  I have.  Things start to break.  Soon you're tracking all sorts of userspace differences, and having to track many other details.  I will not say anything about UEK, but talk to some who have.  And the proof comes out in the immediately following statements.

"The alternative : take a version of Linux that's 3+ years old and take code from a more or less current version, after 100's of 1000's of lines of code changes and architectural changes have happened, and make that new code, somehow work in the old tree."

Yes, sustaining engineering is difficult - rebasing is much easier

Backporting is a masterpiece of sustaining engineering, and a costly one.  But as SuSE first proved in the Linux world by offering a five (5) year release model and Service Level Agreement (SLA) with SuSE Linux Enterprise Server version 7 (SLES7), IHVs, ISVs and Enterprise need it.  Better yet, SuSE proved they were willing to pay for it because they need backports that don't change structures and break their applications.  They need a stable, kernel ABI that mitigates changes.  They need long-term support -- nowdays even 10+ years -- as Red Hat customers have requested.

'The result : old patched up, totally unique code. Someone spending all their time retrofitting complex features into an old version. Testing is unique, bugs found are mostly only relevant in that version and not so much in newer versions unless they're greater design bugs or one liners.'

But this also happens with kernels just a few revisions back.  It happens in Fedora (before Fedora just rebases kernels), and it happens with many others who attempt to maintain a kernel for even just 18-36 months like some "Long Term Support" (LTS) releases.  Red Hat went through such with Red Hat Linux 6.2 Enterprise as well, considered Red Hat's first Enterprise attempt, with only a 3 year SLA, and more rebasing (whereas SuSE offered SLES 7 with a 5 year and less rebasing).  At some point, you might as well know your codebase, which is what Red Hat engineers do.

Red Hat knows the history of the kernel, core C libraries, entire subsystems, even core messaging and desktop interfaces.  They know the history and lineage of the code, and maintain a long swath of it, because customers require such.  If you are a commercial Linux vendor and you do not, then it's difficult to offer an "Enterprise" solution that won't break existing code more than 3 years old, much less 5-7+.  Last time I checked, Red Hat employees now sign off on well over 1/3rd of all kernel.org commits alone -- that's 100% upstream and not related to Fedora either.

It's called stewardship

'Testing is unique, bugs found are mostly only relevant in that version and not so much in newer versions unless they're greater design bugs or one liners. No one else really ever tested it because, well, it's unique, no other company is providing that seriously forked kernel that has 1+M of lines of differences.'

Of course, "Enterprise" releases do have "unique" testing.  IHVs, ISVs and Enterprises rely on that.  They expect it.  In fact, a lot of Linux distributions like to market how quick they are in getting a fix for a CVE out, maybe a few hours ahead of Red Hat and its army of developers and maintainers.  But Red Hat is going to run an entire test suite against it, checking for unit, integration and regression issues.  Red Hat strives to avoid ABI breakage, a major foundation and "value" in its "Enterprise" line of products.

And partners are definitely going to test it.  Testing is not unique to Red Hat at all.  It gets tested downstream as well.  Oh, when Red Hat gets something wrong, the "EL Rebuilds" surely do file the Bugzillas too.  But Red Hat strives to "get it right," whether committing to the upstream kernel.org, to the more leading edge Fedora kernel, or when it comes to backporting and following changes for the Enterprise kernel.  Rebasing is easy.  But IHVs, ISVs and Enterprises don't pay Red Hat to go the easy route, they pay Red Hat to maintain 10+ years of ABI compatibility, or at least mitigate when changes must occur.

How can you integration test something early on, but keep changing things as a matter of policy, without taking on unacceptable risk?  It didn't work when NASA changed the sealant on the SRB O-rings in 1985 with a replacement that was inadequate for moisture and cold, and definitely not after the move to CFC-less EFT insulation in 1997 exposed serious deficiencies in the required tensile strength (sorry, my aerospace background is showing).  Change is to be mitigated in anything that is expected to last 10+ years, because you can never re-integration test to the same lineage of prior testing and resulting perceptions and assumptions.

Perceptions and assumptions that IHVs, ISVs and Enterprises rely on

'Somehow, I prefer our approach. At least the folks at SuSE seem to believe the same.'

My heart goes out to SuSE.  As they claim, they really were the first to prove the "modern Enterprise Linux" model with SuSE Linux Enterprise Server version 7 (SLES).  But now they are smaller, for various reasons I will not go into.  Good people, all good people, many now with Red Hat, among others.  But they cannot keep backporting fixes.  And yes, as a result, even they are even rebasing kernels.  They can no longer maintain codebases and ABIs for 7+ years.  Red Hat is now unique here.

As a developer, I prefer upstream, it's just easier to rebase than backport.  But as a consultant, I have to appreciate backporting and downstream.  Red Hat employees are paid to do both, backport and rebase.  But don't take my view for what it's worth on backporting, take all of the vendors who use and rebuild Red Hat's kernels.

In the virtualization space, both Xen and VMware is still using Red Hat's 2.6.18 kernel releases in both their "standalone" XenServer and ESX products (even version 5).  There are countless others where Red Hat's 2.6.18 and 2.6.32 releases, let alone the fixes they provide upstream for long-term maintenance, have a heavy impact.  This even includes UEK 2.6.32 for OEL5.  You cannot fault Red Hat for its continued dilligence in offering a kernel with 10+ years of backports.  Especially when so many others depend on Red Hat's efforts both upstream and downstream.

Especially anything that is considered "Enterprise."  Red Hat has the Fedora(TM) trademark and associated projects for a reason.

Fedora releases are stable, but they are not "Enterprise" because they rebase as policy

About this Entry
Aug. 5th, 2011 @ 07:53 am Enterprise Linux "Decoder Ring"
Current Mood: accomplished
This is a list of core software versions in Red Hat Enterprise Linux (RHEL) releases that can never be replaced (although concurrent versions, with version appended to name, may be added) and otherwise provides "core ABI/API compatibility" for software long-term. For more information on how Red Hat releases and supports software, see the support article Red Hat Enterprise Linux Life Cycle (updated 2012-Mar).

Modern Linux software compatibility is considered to be software written for LibStdC++ 6 and systems running kernel 2.6 (2004+). Red Hat also maintains some ANSI C++ ABI/API software compatibility with LibStdC++ 5 (GCC 3.3) as well as LibStdC++ 3 (GCC 3) going back a dozen years, longer than any other distribution.

Release
EoL (ELS)
Kernel
GLibC
GCC
LibStdC++
Perl
PHP
Python
Comments
6 2020-Nov (2023-Nov) 2.6.32 2.12 4.4.5 6 (4.4.5) 5.10.1 5.3.3 2.6.6 F12-13
5 2017-Mar (2020-Mar) 2.6.18 2.5 4.1.2 6 (4.1.2) 5.8.8 5.1.6, 5.3.3=php53 2.4.3, 2.6.5=python26 FC6
4 2012-Feb (2015-Feb) 2.6.9 2.3.4 3.4.6 6 (3.4.6) 5.8.5 4.3.9 2.3.4 FC3
3 2010-Oct (2013-Oct) 2.4.21 2.3.2 3.2.3 5 (3.2.3) (3.3-compat) 5.8.0 4.3.2 2.2.3 RHL9
2 (2.1) 2009-May (none) 2.4.9 2.2.4 2.96 2+3 (2.96) (3.0-compat) 5.6.0 4.0.6 1.5.2 RHL7.2
1 (6.2E) 2003-Apr (n/a) 2.2.14 2.1.3 2.91.66 (egcs-1.1.2) none 5.5.3 3.0.15 1.5.2 RHL6.2

RHEL5 Concurrent Versions:

  • php53 - RHEL5.6+ ships a concurrent PHP 5.3 release that is an optional, but conflicting tree with RHEL5 included PHP 5.1 (php)
  • python26 - EPEL5 ships a concurrent Python 2.6 release that is a separate, and concurrent tree with RHEL5 included Python 2.4 (python)
About this Entry
Oct. 23rd, 2009 @ 12:15 am Support 101: Logs and Pastebin
Current Location: United States, Texas, Alvin
Current Mood: awake


I am a member of various communities and I try to do my best to help people solve issues. I'm just a wannabe technologist and solutions provider and do my best to see people through solutions, regardless of distractions such as meta-discussions.

One issue I reguarly run into is that many people are hesitant to send me logs and information so I can pinpoint issues. I don't know if it's really hesitation but more of a matter that they have difficulty following my requests. So, while this might be taken as an insult by some, I'm only providing this as a step-by-step for others. Those who know me know I never talk down to anyone, but only shift down gears to provide complete details for those that honestly haven't done things before.

Hence the reasons for this hastily written "Support 101" entry.


An introduction to GNOME System Log Viewer

I use the GNOME desktop. Your desktop may vary. But for those that use GNOME, or have GNOME installed, there is a helpful, graphical user interface (GUI) log viewer known as, simply, the GNOME System Log Viewer. The location of the Log Viewer will vary based on your site's desktop menu, but by default, it should be under:
  • System -> Administration -> System Log
If you are not a privileged user (and you should not be, this isn't Windows where programs break if they aren't running as a privileged user), then you will most likely be prompted for the password of the "root" user. Figure 1 illustrates a typical prompt by the GNOME desktop.

gnome_query

Figure 1
: Administrative Privilege Query

If your distribution does not allow you to enter root credentials directly, then it may require use of the "sudo" command. In that case, open a command line interface (CLI) prompt and use the "sudo" command to execute the "gnome-system-log" command. E.g.,
  $ sudo gnome-system-log
  Password:  
You may have seen some uses of sudo to launch a full, privileged shell, such as illustrated in Figure 2.

gnome_cli

Figure 2: Full Administrative Privilege Shell in a typical Linux Command Line Interface (CLI) via Sudo

In any case, once the System Log Viewer launches. It shows a three (3) pane display of ...
  • List -- list of common files with relevant system information
  • Calendar -- when not grayed out, bolds the dates of entries in the current log view
  • View -- the contents of the log file
The main log of any POSIX system is its primary Syslog output file, which is /var/log/messages on Linux systems. Like most entries in the System Log Viewer, entries will be parsed by date and can be "folded" and "expanded" to hepl with organization. If you're like me, with a notebook computer, you're often booting every day. So the first entries for any day will often be the initial boot message when the system was first booted that day, as illustrated in Figure 3.

gnome_log_main

Figure 3: Main Syslog entries for Thursday, October 22nd, starting when the system was first booted at 1:46am.

Of special note is that fact that some logs are "rotated." E.g., in the case of /var/log/messages, older version of the log are named /var/log/messages.1, /var/log/messages.2, etc... This is the result of the typical logrotate configurations on most Linux systems. The System Log Viewer notes this and will list older versions in a pull down in the lower right corner, as illustrated in Figure 4. When accessed, the calendar will even bold the appropriate dates with entries in these older versions too, adding additional, visual information for quick notification.

gnome_log_version

Figure 4: The lower right pull-down menu with older log versions in the System Log Viewer.

Another important log is the X-window version 11 (X11) logs, typically the implementation from The X.org Foundation, and its /var/log/Xorg.0.log output. In this case, as illustrated in Figure 5, View -> Filter (also Ctrl-F key combination) was used to filter out all but lines with "DDC" (VESA Display Data Channel).

gnome_log_ddc

Figure 5: Filtering out lines by keywords in the System Log Viewer.


Directing output into a new/custom "log" file

Sometimes information will not be located in a log file or otherwise needs to be captured into a file or other form for use later. POSIX systems can utilize (as well as Windows, being that MS DOS 2.0 and later directly used SCO Xenix code to add) redirection of output to a file. Figure 6 illustrates redirection of the output of "lspci" into a file, and then the GNOME Editor is launched to view the file.
  $ lspci > /tmp/lspci ; gedit /tmp/lscpi &

gnome_cli_gedit

Figure 6: Redirecting command output to a file and viewing the file in the GNOME Editor.

Of course, this new file (/tmp/lspci) can be viewed in the System Log Viewer as well. It's as simple as using the menu:
  • File -> Open
  • Browse to file location (Filesystem -> /tmp)
  • Pick the file
Figure 7 illustrates the lspci output now viewed in the System Log Viewer, with all lines highlighted by clicking on the first line, then holding down the Shift key and clicking on the last line.

gnome_log_lspci

Figure 6: Viewing other/custom files in the System Log Viewer (with all lines highlighted in the view pane)


Uploading file contents to the Internet via Pastebin

Most forums have limited file attachment and/or viewing capabilities. When monospace and other attributes should to be retained in the original format, solutions like Pastebin should be utilized. Pastebin is a simple PHP program which allows file contents to be posted on the Internet and viewed, including options for formatting (such as programming language-specific syntax formatting). There is a public Pastebin.COM site where anyone can upload file contents, although organizations with more confidential information should consider setting up an internal server (only accessible via login, VPN, etc...).

After highlighting the contents of the custom lspci output file in Figure 6, the Edit -> Copy (also Ctrl-C) menu can be used to copy the contents of the file to the clipboard. Once copied to the clipboard, the contents can be pasted into a browser with Edit -> Paste (also Ctrl-V), such as one opened to Pastebin.COM as illustrated in Figure 7.

gnome_pastebin_paste

Figure 7: Pasting file contents into Pastebin

Pastebin (plus various Pastebin/PHP/Apache formatting extensions) supports many different syntax formatting options. If none are appropriate for the contents, the double at symbol (@@) can still be used as the preface for any line, which will highlight the entire line. As entered in Figure 7, and illustrated in the resulting output of Figure 8, the third (3rd) line is highlighted, which shows the display adapter in the lspci listing.

Note the resulting URL in the address line of the Browser of Figure 8. This is the URL to share the link to others in an e-mail or forum posting, so they can view the contents of the file. Also keep track of the link and your declared "Name" so it can be edited or deleted (browser settings or other session state permitting).

gnome_pastebin_sent

Figure 8: Pastebin output with custom URL and syntax formatting


I will be linking to this blog entry in the future.  I sincerely no one takes it as an insult and only as someone who honestly wants to help, especially getting those all-important logs and various command out that will tell me exactly where your issues lie.

About this Entry
Aug. 27th, 2009 @ 10:11 pm Legacy and Incompatible PC Power Supply Connectors
Current Location: United States, Maryland, Glen Burnie
Current Mood: tiredtired
Tags: , , , , , , , , , , , , , , , , ,

If there is nothing that drives me up the wall more than anything else, it's commodity PC vendors reusing existing connectors for new, incompatible, allegedly (often hacked, then adopted) standards.  And there is nothing more of an issue in this space than power, where you can fly or fry very easily.  In fact, keyed connectors are often useless, as they do not prevent improper voltage.

Most of the details I'm going to cover are not well known and probably don't need to be known in 2009.  However, in case you run into legacy systems with incompatible PC power connectors, be aware of these possibilities.  I originally blogged a warning about the 2x3 (6-pin) conflicts back in 2005, but I figured it's time to put this all in one article.

Know thy basic colors

When it comes to power in PC systems, although you cannot trust or guarantee these to be true, they are typically true in the 21st century at least, with Black being utilized for most GND (although not necessarily negative voltage).

Color
Voltage
Yellow+12V
Red+5V
Orange+3.3V

Today's Auxillary PC Power Connectors

For the most part, virtually all of today's auxillary PC power connectors are +12V-GND pairs.  These aux connectors include ...

"P4" Connector
(typical ATX12V and ATX v2.0+)
GND+12V
GND+12V
2x2 (4-pin) configuration

SSI EEB "Server"
(EPS12V w/"Server")
GND+12V
GND+12V
GND+12V
GND+12V
4x2 (8-pin) configuration

PCI Express (PCIe) Power
(ATX12V and EPS12V)
+12VGND
+12VGND
+12VGND
3x2 (6-pin) configuration

Now to confuse things even more, some PCIe connectors have a 4x2 (8-pin) configuration, which merely adds two (2) more GND conductors.  Luckily the industry "non-standard" vendors (who do their "own thing") were smart enough not to make the connector shape exactly the same as the SSI EEB Server connector.  So unless you force it, you can't accidentally plug one into another.

However, as noted in the next section, the PCIe power connector actually did make a major mistake with another SSI EEB standard.

SSI EEB "Workstation" (2002-2005)

To support new, standard workstation mainboards with additional +3.3V requirements, the Server System Infrastructure (SSI) forum created another power connector for "workstations," that supplemented the 4x2 +12V "Server" connector.  This was known as the "Workstation" or "WS" connector.

SSI EEB "Workstation" (WS)
(EPS12V w/"Workstation")
+12V+12V
+3.3VGND
+3.3VGND
3x2 (6-pin) configuration

This connector was utilized by most Athlon MP implementations, as AMD pushed for a single, open power supply standard instead of various vendors with their own, proprietary ones.  The SSI EEB "Workstation" (WS) replaced the stop-gap ATX GES (discussed later).

Now here's to the brilliance of the "non-standard" industry vendors.  They not only created the PCIe power connector using the same 3x2 (6-pin) configuration.  The PCIe vendors irresponsibly used the exact same keying as SSI EEB "Workstation"!  Brilliant!  (NOT!)  The SSI EEB "Workstation" connector very much pre-dates the introduction of PCIe video cards and their need for additional power.

When this happened, AMD and major, Tier-2 Whitebox OEMs (like SuperMicro and Tyan) had to issue technical documents warning people not to use PCIe power connectors.  Here's an example for the Tyan S2895 warning against plugging a very common PCIe power connector into this board (and note that exact same keying as the SSI EEB "Workstation" spec).  Likewise, those who purchased EPS12V power supplies in the 2002-2004 timeframe may have a 6-pin connector with two (2) Red (+5V) wires.  This is the "Workstation" connector, not the PCIe.  Don't plug it into your PCI card either.

AMD GES Power Supply (2000-2002)

When AMD brought the Athlon MP to the market, there was no open standard power specification that could handle its multi-socket power requirements.  This was before the SSI forum, SSI EEB, and virtually all vendors had their own, different, proprietary specifications for multi-socket intel solutions.  So AMD introduced it's own, GES.

Although AMD GES is often referred to as ATX GES, it is not remotely ATX compatible.  It was purposely not so, with different keying as well.  At the time, ATX was still a 2x10 (20-pin) P1 connector.  AMD GES introduced a new 2x12 (24-pin) P1 connector, but also with different keying.  The SSI EEB would later create a new 2x12 (24-pin) EPS12V connector which largely has a P1 compatible with ATX12V (anything ATX v2.0+ typically works without issue, although not all ATX12V based on ATX v1.0).

AMD GES "P2" Power
(short-lived AMD GES)
+12VGND
+12VGND
+12VPWR-OK
+3.3VGND
4x2 (8-pin) configuration

AMD GES also has a 2x4 (8-pin) 'P2" connector, but it is not compatible with the latter 2x4 (8-pin) SSI EEB "Server" connector.  So instead of inspecting the P1 connector, the easiest way to identify an AMD GES power supply is via the 2x4 (8-pin).  It'll have more than just yellow (+12V) and black (GND), but one Red (+5V) wire.  It should also have a Grey (PWR-OK) wire as well, instead of a GND.

Rackmount.COM has a great dissection of all the differences between AMD GES and EPS12V (as well as older ATX v1.0, prior to the ATX v2.0 alignment of ATX12V and EPS12V).  They also sell converter cables so one can use a newer EPS12V supply for AMD GES mainboards.  I have personally utilized these cables and they are much cheaper than trying to find the few AMD GES power supplies out there (typically 460W).  The creation of the SSI forum and EPS12V standard removed the need for AMD GES, and one can even see the influence on the "Workstation" connector from it.

Just some connectors to recognize in case you run into older systems.

About this Entry
Aug. 20th, 2009 @ 09:46 pm Two (2) Per-user Methods to Disable PackageKit Notifications in GNOME ...
Current Location: United States, Maryland, Glen Burnie
Current Mood: exhaustedexhausted

I'm going to make a quick blog of this.  I'm not going to debate the merits and purpose of PackageKit, how it operates, different distro defaults or other meta-discussions on the matter.  There are decisions and reasons that have been discussed enough, let alone related discussions (e.g., the common "why did Yum take a different approach than proven/assumed Apt operation?").

PackageKit, like many solutions out of Raleigh-Westford-et al. (i.e., largely Red Hat), are designed with GNOME in-mind, but there are icons, launchers and interfaces and other support in KDE on Fedora-based distros, among others.  KDE references can be found elsewhere.  The following are the per-user settings that you can change in the GNOME desktop GUI in seconds.

Option 1:  Disable All PackageKit Operations in Future GNOME Sessions for the Current User
  • Either:  From GNOME menu System
    • Select Preferences > Software Updates, or
  • Or:  In the notification area (defaults to right side of the bar near the clock)
    • Right-click notification icon for PackageKit applet (typical mouse-over is "There are # updates available")
      • Select Preferences
  • Now in window Software Update Preferences
    • In pull-down options next to Check for updates:
      • Select Never
    • In pull-down options next to Automatically install:
      • Select Nothing (should be the default)
    • In pull-down options next to Check for major upgrades:
      • Select Never
    • Click button Close
Option 2:  Disable Startup of PackageKit Applet in Future GNOME Sessions for the Current User
  • From GNOME menu System
    • Select Preferences > Startup Applications
  • Now in window Startup Application Preferences
    • In tab Startup Programs
      • Highlight PackageKit Applet (scroll down using the scrollbar on the right if needbe)
      • Either:  Uncheck the checkbox
      • Or:  Click button Remove
    • Click button Close
In both cases, the setting is per-user only.  If the system-wide service daemon /usr/sbin/packagekitd executes on the system, it will continue to execute after these changes.  However, they will modify how the currently logged in user executes the user's PackageKit Applet.  And in both cases, the change in settings will not take effect for the current user until the GNOME session is ended and another one started (e.g., the user logs out and back in).

With Option 1, the PackageKit Applet still executes, but the notification icon should not appear with one caveat.  PackageKit will no longer check for any updates for the current user.  However, in the case the system has not been updated since the last time PackageKit did check for updates (before the changes were actually made), the notification of updates may still appear.  Updating the system will remove this notification, and it will never return for that user.

With Option 2, the PackageKit Applet is never launched when GNOME starts for the user.  So the notification never appears in the next GNOME session, regardless of the prior state of updates.

I will leave all other debate, discussion and consideration to other articles, including system-wide changes -- including just uninstalling PackageKit.  However, this per-user method and setting is the only way to guarantee that the setting for users does not change, regardless of what a system administrator or other privileged user or related operation does to affect PackageKit at the system-level.  E.g., PackageKit may be re-installed, but the user settings will not change.

NOTE:  To make these GNOME settings the default, mandatory or on a per-user basis outside of the GUI when the user is logged in, see the GNOME Desktop System Administration Guide.  Yes, GNOME has full policy capability -- mandatory, default, per-user, etc..., as does anything Mozilla (e.g., Firefox, Thunderbird, etc...).  I regularly have to educate even experienced Linux professionals that, yes, you can lock-down Linux desktops quite well, with both a centralized hierarchy and utilizing a tier, contextual approach (as simple or as complex as you want to make it).  There is even a costly, per-user solution to add this management capability to Active Directory's MMCs and GPO facilities, instead of using file-based or open source LDAP-based (e.g., Port389.org, Red Hat Directory Server, etc...) stores.

About this Entry
Jul. 21st, 2009 @ 10:00 pm Subnetting and CIDR-to-Mask "Check" Sheet
Current Location: Bowie, MD
Current Mood: awake

I'm not big into "cheet sheets."  As my Introduction to Environmental Engineering instructor utilized as a most excellent approach (beyond his first day, "if you haven't had Calc II and Mechanics/Statics, get out"), "I'm going to give a sheet with all the formulas you need for the exam, plus all sorts of other formulas from aerospace, civil, computer, electrical and mechanical engineering, maybe a few ones from physics too."

In other words, you don't need to memorize, but you do need to understand what you're looking at.  Especially in engineering, which is the study of systems, building a system of equations to explain systems, etc... and there's a lot of overlap (e.g., mechanical "normal" v. electrical "right-hand-rule").  We don't memorize in engineering, we derive and then reference work that has already be learned.  And most importantly, we check our work, we are methodical, not necessarily smart.  In fact, it's the fact that we recognize we aren't smart and often wrong that makes us engineers (sans the one gal/guy that was a 4.0 student that blew the curve every time -- don't count her/him).

So, with that said, I will offer this "Check Sheet" for Subnetting and CIDR-to-Mask conversion.

Why do I call it a "Check Sheet"?  Well, two reasons ...
  1. You should be able to create it from memory, and
  2. I make mistakes, so I should look to it as a check/reference when pressed for time
I've done enough IPv4 (and even some IPv6 if you can believe that) that I know CIDR prefixes and Subnet masks in and out.  But sometimes I still make mistakes.  Like my Electrical Networks (i.e., Analog Circuits) instructor, not merely a PhD but co-chair of the laser research center and one of the smarted individuals I've ever met, used to make simple arithmetic mistakes after deriving a complex series of equations only half of us followed, I make some stupid ones at times.  I could care less about all the theory, like I did with astrophysics -- I just want to the applied engineering, like I did with engineering mechanics.

Technical application is more of the same from the "theory" of engineering technology.

So, without further delay, here's my "check sheet."


When you take a networking exam (CCNA, JNCIA, etc...), learn to write these down on your scratch paper.  You should be able to create these from your mind without much effort.  These "reference tables" will save you a lot of time and effort when answering questions that, well, are just dead simple.

Table 1 (Optional):  Know your Base 2 (binary) powers to 10 ...
2^01
2^12
2^24
2^38
2^416
2^532
2^664
2^7128
2^8256
2^9512
2^101,024 (Ki)

Now why do you only need to memorize 0-10?  Simple.  Because 2^10 is 1,024, or one Kilo-binary (KiBi, Ki for short).  Take 100,000 for example.  What powers of 2 are greater than 100,000?  17+.  You know this because 2^7 = 128, add 10 for the 1,000.  In all honesty, you should memorize these over time and not make a mistake at all.

Table 2 (Recommended):  Know the CIDR Prefix to Subnet Mask Octet (one of the four numbers in a IPv4 address)
CIDR Prefix
Mask Octet
/0, /8, /16, /24
0 (256)
/1, /9, /17, /25
128
/2, /10, /18, /26
192
/3, /11, /19, /27
224
/4, /12, /20, /28
240
/5, /13, /21, /29
248
/6, /14, /22, /30
252
/7, /15, /23, /31
254
/8, /16, /24, /32
255

Yes, there are 32 possible CIDR Prefixes / Subnet Masks in IPv4.  But in reality, you only need to know a quarter of them, the individual octet. This table you should write down and reference as a "sanity check," because -- unlike powers of 2 -- sometimes you forget.

Start with /0 and then add the perfect classes A (/8), B (/16), C (/24) in the left column and write "0 (256)" in the right column.  Now write the next iteration of each (/1, /9, /17, /25) and the half-way point of 0 and 256 = 128.  Now write the next iteration and the half-way point between 128 and 256.  So forth.  You should learn all of the mask octets over time, but even then, sometimes it's best to have the table for quick reference.  On more than one occasion I've transposed the octect (e.g., 254 for /17, instead of 128).

Why did I write /0 out?  If you didn't notice, there are 4 values in the left column.  If you have /22, then you know it's the third position and the octets to its left are 255 and the octet to its right is 0, the 252 goes in the third position, just where /22 is located -- 255.255.252.0.

So what do these two tables give you?

CIDR is the most useful because it's powers of two.  Mathematically CIDR /18 means 256 Ki (~ 256 thousand) networks of 16 Ki (~ 16 thousand) host.  I'm not bothering with the math any more than I would want an universal equation for astrophysics, give me a real application like orbital mechanics!

You have the first table, plus you know IPv4 is 32-bit (32 powers of 2).  18 = 10 + 8 => 1 Ki * 256 = 256 Ki > 256,000 (actually 262,144, yes, memorized that long ago).  How many hosts?  Remember that 32?  Yeah, without any subnetting, you have 2^32 hosts, so you need to subtract the CIDR (networks) from that 32 to get the power to use for hosts.  32 - 18 (networks) = 14 (hosts) = 10 + 4 => 1 Ki * 16 = 16 Ki > 16,000 (actually 16,384, which you may have memorized too).

Now you have the fomulas for taking X networks v. Y hosts and subnetting/supernetting in CIDR, given X networks and/or Y hosts.  Subnetting increases CIDR (more networks, fewer hosts), supernetting decreases CIDR (fewer networks, more hosts).  It's typically best to start at a Class A (/8), B (/16) or C (/24) and increase the CIDR (more networks, fewer hosts).

But sometimes questions like to "trip you up" and have you supernet set of a Class Cs.  Just remember that a supernet of Class C (/24) networks is a subnet of a Class B (/16), so the CIDR should head towards B (/16), less than C (/23).  No difference, just remember that the whole idea of "Classes" is so '80s ("And I Ran ... Ran So Far Away" ... yeah, we all did from the '80s) and don't let the "Class" confuse you.  Just remember the "perfect 0" in the octet exists on A (/8), B (/16) and C (/24) and is a good "starting point."

Adding in the second table, you now have a way to convert to/from CIDR prefixes into subnets rather easily.  Line up the CIDR in the left column as 2nd, 3rd or 4th octet, put 255 in the ones to the left, 0 in the one to the right (if applicable), and then the value of the right column as in the same octect position in the left column for the CIDR prefix.

Practice using the binary (through the 10th power) and the CIDR to netmask octet (only 8 to create) tables.  You can probably drop the first table after awhile (3 -> 8, 4 -> 16, 5-> 32, etc... should be memorized after awhile), but use the second table as your "sanity check" for CIDR to/from Netmask.  Work in CIDR as its the easiest, everything is between 0 and 32.
About this Entry
Jul. 2nd, 2009 @ 11:41 pm Microsoft Certified IT Professional (MCITP) on Windows Server 2008
Current Location: Glen Burnie, MD
Current Mood: awake

2009 June:  An "Exciting" Month (meaning little sleep)

June was an "exciting" month.  By "exciting" I mean I was so busy with professional details (much less personal) that I slept a median of three (3) hours an evening.  It started on Monday, June 1st around lunch when one of my bosses informed me my initial "risk assessment" (not my place to say more) so I could start at federal client next Monday.  That meant I had 48 hours to pack up my "mobile office" (as I call it) for temporary relocation so I drive up Thursday and arrive by Friday to be badged and sworn in.  I spent the weekend looking for housing and was slammed with the job and related, relocation administriva over the first three (3) weeks of the month.

June was also the month I scheduled four (4) new Microsoft certification exams for myself, and was required to take them by June 30th.

It started in late April with an offer from Prometric.  For those of us Microsoft Certified Professionals (MCP) who had not sat an exam since prior to 2007, a coupon was offered for $100 off of the exam fee.  Microsoft exams are typically only $125, so this meant the cost was only $25 and nearly free.  Even better was the lack of limits on how many times the coupon could be applied.  The only catch was that the exams had to be sat by June 30th.

So timing was everything here.  I signed up for a full Microsoft track, plus one additional exam down another.  Being slammed with temporary relocation and a new job, I was rushed to take the four (4) exams in the final week of June (23rd - 30th), while working handling other details.

The Microsoft Certified IT Professional (MCITP) Program

The one track I signed up for was part of the not entirely new Microsoft Certified IT Professional (MCITP) program.  The MCITP replaces several "system administration" and "technology" certifications, unifying them under a single acronym.  The MCITP designation actually began after SQL Server 2005 was released, and now encompasses Windows Server 2008 and other products.  Specifically, the following programs are replaced: 
  • Microsoft Certified Systems Administrator (MCSA)
  • Microsoft Certified Systems Engineer (MCSE)
  • Microsoft Certified Data Base Administrator (MCDBA)
  • Microsoft Certified Desktop Support Technician (MCDST)
In fact, the Microsoft certification programs have been separated into five (5) separate foci:
  • IT Professional (MCITP) replacing the separate MCSA, MCSE, MCDBA and MCDST
  • Professional Developer (MCPD) replacing the separate MCAD and MCSD
  • "Advanced," the new Master (MCM) and Architect (MCA) programs (the latter being a 2-week, on-campus engagement)
  • Application Specialist (MCAS) for Home and Office replacing Office Specialist (MOS)
  • Trainer (MCT) and the Learning Consultant (MCLC)
The administrator/technologist MCITP program also introduces the indivudal Microsoft Certified Technical Specialist (MCTS) title for supporting exams that feed into MCITP titles.  The titles of the exams themselves are prefixed as such with "TS", with the final, qualifying, "PRO" exam for each program.  E.g., several TS exams may be required and shared between different MCITP tracks, but then there is at least one, dedicated "PRO" exam finale for the MCITP track.

For those that are traditional engineers and engineering technologists, this means an end to the use of the term "engineer" (which may be reserved in various locales) for Microsoft.  Of course, traditional architects may still have a complaint (in some locales).  As a traditionally educated engineer with some software engineering experience, I thought this was of note, and I have to commend Microsoft for removing the long-standing issue of "engineer" v. "technology" in titles.

MCITP: Server Administrator and Enterprise Administrator

Microsoft has replaced the MCSA and MCSE with a pair of tracks for Windows Server 2008, although Microsoft's FAQ states that the pair are not replacing the prior MCSA and MCSE programs 1:1.  Probably one thing to point out, which probably clarifies this, is that you can only upgrade from the MCSE on Windows Server 2003.  There is no upgrade path for the MCSA on Windows Servrer 2003.

MCITP "Administrator Common" (2 exams)The 070-640, now retired and replaced by the 083-640 in the US, and the 070-642 represent the "core" Technical Specialties (TS) of the new MCITP program for System and Enterprise administration.  The 083-640 exam, introduced 6 months earlier, has "performance-based components" -- the first ever by Microsoft.  These are not unlike the "practicum" approaches in the Novell NCLP and CLE programs where you connect to a remote, live, real server.  This is not the same as the "simulation" portions of select, other Microsoft exams.  Microsoft has stated that it plans to add more 083 exams to replace 070 equivalents.  Sadly, only the latter digits show up on your transcript (no proof you actually sat the exams with "performance-based" components).

I talk more about my "experience" with the 083-640 later in this entry.

UPDATE:  By 2010, after one (1) year of countless crashes and other issues, I'll speculate to say using the clearly non-native Linux format also wasn't winning any converts, Microsoft has dropped the 83-640 and returned to the traditional, CBT 70-640 approach.  I have yet to see another 83 exam offered in any track.


MCITP Server Administrator (+1 exam)MCITP Enterprise Administrator (+3 exams)And here are the two levels of the new MCITP for system and enterprise administrators.  As you'll note from the above, the "Enterprise Administrator" adds detailed Client (Vista/Office) and Application (IIS/Services) over the "Server Administrator," along with a different "PRO" exam covering the wrap-up of those added details.  While the "Server Administrator" does dive into some client and application details, including design in the "PRO" that touches on both.

Regarding upgrades, Microsoft offers no Windows 2000 upgrade track, and no MCSA on Windows Server 2003 upgrade track.  Only those MCSEs on Windows Server 2003.  And unlike prior upgrades from NT 4.0 to 2000 and 2000 to 2003, the upgrade exam does not reward the new certification.  The "upgrade" exam only replaces 070-640, 070-642 and 070-643, the TS "Server" exams, but neither the TS "Client" nor the PRO exams.  The resulting track then becomes ...

MCSE on 2003 (only) "Upgrade" to MCITP Enterprise Administator (3 exams total): 
The 070-649 exam replaces the 070-640, 642 and 643 exams.  In other words, you're only "upgrading" to select, individual "Technology Specilist" titles on 2008, not the full MCITP title.  Although there is no MCSA path, I guess it's possible to "upgrade" to the MCITP: Server Adminstrator by taking 070-646 "PRO" with the 070-649, but then you're talking two (2) of three (3) exams, and the upgrade only saves you one (1) exam for the "Enterprise Administrator."  In any case, it's clear that Microsoft is trying to differentiate the MCITP from the MCSA/MCSE prior.  I just don't know what they expect people to think.

My Experiences With the New MCITP Program

As I mentioned before, I originally scheduled the exams with Prometric for Florida in May.  I was going to sit the exams on Saturdays in June.  Unfortunately, the relocation changed that, and I actually had to call them to reschedule for a different state.  Prometric's web-based interface does not let you reschedule along state lines, which has been a major issue for myself in the past (and even caused me to forfeit some exams and have some issues with Cisco on the timing of my CCDP).  The administriva kept causing me to push back my exams until the very last week in June, where I finally just ignored sleep for study and sat them.

Originally I scheduled 070-640, 642 and 646 in the "Server Administrator" track, and then the 070-620 Vista Client to start the "Enterprise Administrator" track.  I figured I'd work on the "Enterprise Administrator" later in the year.  In any case the $400 savings on four (4) exams made it worth it.  When I reschedule my exams from Florida to Maryland at the start of June, I found out that the 070-640 could not be rescheduled.  After the Prometric representative dealt with the extensive administriva on the matter, she informed me I would have to sit the new 083-640 version with performance-based components.  I was actually excited to finally see Microsoft adding such.

Timing with work finally led me to schedule the following exams on these days, literally the last week of the month, as the Prometric coupon required sitting (or forfeit of the registration) by June 30th (although it said the registration was good for a year under the confirmation number, it would not let me schedule at any site beyond June 30th).
  • 070-646 on Tuesday, June 23rd -- my 12th ever Microsoft exam
  • 070-620 on Friday, June 26th -- my 13th
  • 083-640 on Monday, June 29th -- my 14th
  • 070-642 on Tuesday, June 30th -- my 15th
I assumed the "PRO" exams would be easiest, so I scheduled the 070-646 first.  Some might disagree, but from my prior experience on my MCSA/MCSE, the "design" exams are always easiest in my view.  Why?  They test layout, integration and other things of various Microsoft services and support.  If you really understand Microsoft's architecture, which you should if you've been working on NT since the original 3.1 beta -- much less the decisions and differences from OS/2 and VMS prior, and the Michigan LDAP / Netscape iPlanet, MIT Kerberos and other "code grab" that became Active Directory -- it's not difficult.  I only had to learn the new services and where
they fit in the "Microsoft Solutions Architecture."

Sitting the 070-646 and 070-620 exams

Although I had been studying Windows Server in general whenever I could on my 17" notebook (I love having a true desktop replacement with dual-500GB drives and 4GiB of DDR3 memory so I can run a lot of virtual machines), I had stupidly studied the 070-642 materials instead of the 070-646 exam the week prior.  I realized this just before lunchtime on the 23rd itself, the day of the exam, so I added a few PTO hours to cram from the 646 manual.  Thank God I downloaded the CDs from the Microsoft Study Kit, including the PDF version of the book.

I passed 070-646 with over a 900 out of 1000 (933).  I have yet to not pass a Microsoft "design" type exam without coming with 7-8% of perfect.  Despite not giving the 646 my full attention, it didn't make much difference.

In all honesty, and this is on me, I "blew off" studying for the Vista Client in 070-620.  My attitude started when the Virtual PC 2005 image of Vista Enterprise would not work under Virtual PC 2007.  So between work and preparing for the 083-640 on Active Directory with the new "performance-based" components, I only gave Vista marginal study without hands-on.  In all honesty, I don't use Vista much.  It's only on my personal notebook, it's the "Home Edition" anyway, hence why I wished the Vista Enterprise VHD would have worked.  That's Microsoft for you.  I use Linux desktops far more at work and home and at a client, if I have to use Windows, it's always XP Professional (let alone I'm largely running from Cygwin under it, and using open source software).

When I sat the 070-620, anything that was Server interaction related, I slammed dunked.  I did know my remote access and wireless components as well.  But the multimedia and other "consumer" components, I bombed ... badly!  The score sheet's "graphical breakdown" showed where, well under 50% in those portions.  Yes, I don't use Vista at home at all for multimedia and "consumer" stuff, only Linux (and that's been the case for the past 12 years).  And that cost me, as I scored only 644/1000, where 700 was passing.  It was my 13th Microsoft exam, the only the second time I didn't score at least an 82% overall.  While I squeaked by on the 070-214 MCSA Security exam without study with a 783 or so, just over the minimal passing, back in 2003, this time I clearly failed by 5.6%.

It was the first Micrsooft I had ever failed, my 13th ever Microsoft exam.  Thirteen is supposed to be my lucky number, as I was born on Friday the 13th.  Guess it really is the opposite when you pass all the others.  ;)

Welcome to 083-640 and the "Lost Connection"

Putting that aside, I focused on studying for the 083-640 throughout the weekend.  From the 646 exam I already knew about the "Domain Level" differences between 2003 and 2008 (I already knew 2000 v. 2003 from experience), and laughed at the fact that the "Forrest Level" differences between 2003 and 2008 are nothing (I honestly sure wish I knew when talking to one client who said they couldn't install Red Hat Directory Server and the AD sync because they use 2008, a domain can be 2003 in a 2008 forrest and lose *0* functionality).  I played with the GPO changes and laughed at various details about Security Objects that require "shadow groups" (defeating the purpose of the option given limitations in AD's OUs and other things -- hack after hack after hack, that's what the test is about!).  I felt ready.

When I sat the 083-640 exam, it was given in three (3) parts:
  • 65 minute lab 1
  • 65 minute lab 2
  • 60 minute, traditional 30 Q&A
Each lab cannot be returned to after you hit "Done."  You are given tasks in each lab, you are on a physical system and can do them in any order (just like the Red Hat exams) and when you hit "Done," you're done with it (Red Hat exams reboot the system, so the system has to be "as is from a clean boot" when scored).

What I saw before and when I entered the first lab shocked me.  First I got a pop-up from the local program telling me that I needed to click a checkbox and accept when a security window popped up.  Sure enough, I got a NTLM -- yes, 15 year-old NT Lan Manager (NTLM) security pop-up -- saying that I trusted the connection, overriding the default because it was, in fact, not trusted.  With all the study I had put into, all while digesting all of the Microsoft marketing on, Read Only Domain Controllers (RODC), Lightweight Directory Services (LDS) and, most notably, Federated Services (FS) -- things that were designed to allow AD to "feed" things and "be more secure," all without opening up AD itself with all its RPC and other services, using secure links, modern hashes, etc... -- I got a NTLM prompt!

And then it got comical.

The screen came up.  It had serious latency.  Not only that, the "pick" looked familiar, darn familiar.  No, that's not Terminal Services pick in RDP.  No, that's not Citrix's XenApp ICA protocol and presentation either.  And then the repainting of the screen told me the truth.  Yes, that's framebuffer, dumb framebuffer, and the all-too-familiar VNC pick!  Oh yeah, this isn't a Microsoft solution.  It's a 3rd party, pass-thru NTLM authentication, VNC framebuffer based solution.

And that's where the frustration kicked in.

Normally, the tasks they give you here would only take 10 minutes, possibly only 5, if you knew what you were doing.  If you have to look through a few minute items or launch a couple of programs or -- even more so -- expand those detailed policies to find the few policy objects you want to modify, but don't remember the exact hiearchy to get there, maybe 10-15 minutes on a local system.  But no, were remote, major latency, and the screen paints like garbage.  That's why they give you 60 minutes.

It gets better ... "lost connection."

I made it through the first lab, probably a little frustrated between the lack of sleep, general dislike of the format and disappointment.  This was especially the case after taking so many Red Hat exams on physical, local hardware, or at least having a "command line" that works well over a VPN (and such is the case in the EX436, clustering, where you have a half-dozen systems, and some operations cannot be done on the 1-2 physical, local systems).  I would later hit myself in the head when I realized where I should have gone in the menu.  The darn latency kept me from exploring options, and I know I configured 2 things wrong that I should not have, and would have easily discovered had I been able to "play" on a local system.  But then I hit the second lab.

After the first 20 minutes into the second lab, it "lost connection."  It sat there with a green silon type back-forth, back-forth trying to re-establish connection.  All-the-meanwhile the locally executing, right-hand side of the screen kept counting down both the lab time remaining, as well as entire exam time remaining.  After 5 minutes I notified the local proctor.  He was very nice (I had visited the same location before), but it took me a good 3-4 minutes to get him to realize that this new, first of its kind (exact words from Microsoft's blogs) 083 exams were using a remote system.  So he contacted his Prometric/Microsoft people.

After answering various questions of when I started the exam -- the remote servrer is tied to its windows from various locations -- they had him reboot the local computer.  Because of the same reality, that multiple testing locations are hitting the same, remote server and the access slots are "windowed," I lost those 15 minutes!  What was around 40 minutes remaining became 25 minutes, and there was nothing the Proctor could do about it.  You'd figure they'd "buffer" in some extra time per session for these types of situations, but they do not.  And so I re-entered the exam with about 23 minutes remaining.  I tried to focus on just continuing on, but after 2 minutes, I just hit "Done" in disgust.

The traditional section then came up and I went through it very fast.  It was typical Microsoft, a really "great answer" with "one little thing wrong" and then a "poor answer" but "the only one that worked."  Typical.  I didn't check my answers.  At this point I was just frustrated.  I kept my wits about me.  In fact, when the Proctor said, "well, what can I say, it's Windows," it did make me laugh.  It certainly wasn't his fault and I just wanted to try to "give it my best" to finish.  But I wasn't going to spend much time checking my work, and I was even debating if I wanted to sit this exam again.  I'd much rather have only 15 minutes per lab on a local system than 65 connecting over the Internet, but alas, this is the reality of the format that Microsoft is pushing.

I hit my final "Exit Review" and got my result.  No way, I passed with a 783/1000 (700 was required).  In all honesty, I "knew what I was doing" in the virtual labs.  Although one cannot talk about the exam, I knew how to do 100% of the lab content, and it was just a matter of remembering where that one box in that one tab under that one MMC pane was (if in a MMC) -- something that is not easily done in such a remote environment over an Internet connection without frustration.

When the "evaluation/comment" questions came up, they hit the point home.  It explicitly asked about the lab v. traditional portions, including any technical difficulties (including "crashes"), interactive feel, etc...  I know Microsoft beta'd this test for 6 months (as 070-117 IIRC) before starting to offer it 6 months earlier before making it mandatory in the US starting in June.  But this is a really poor setup, and I assume it's leveraging the same Prometric/Novell engine.  In fact, on Tuesday, I meet a former Novell-SuSE employee with a NCLP+CLE at my clients HQ office and we talked about this.

Although he had never sat 083-640, he said the NCLP+CLE use the same.  More surprising to me, he commented how he loves the RHCE format much better, with a physical system, and the fact that Red Hat can cram in a lot more "real testing" in 3 hours than the Prometric remote connection could allow in 12.  Nice to hear that "independent, unbiased" viewpoint, let alone from a former employee of a Red Hat competitor who has sat the competiting Linux certification program.

Once again, I want to stree that the 083-640 and its "performance-based components" are not the same as the "simulations" on some Microsoft exams.  In the "simulations," they are running local, only take 5 seconds to launch, and you can't click outside of select windows and bars.  With the 083 series "performance-based components" of remote labs, you have a full, real system running at the other end (Microsoft stated the system is running as a virtual guest), and you can do anything on them.  That's the easiest way to compare.

UPDATE:  Again, by 2010, Microsoft has dropped the 83-640 and returned to the traditional, CBT 70-640 approach.


Finishing off with the 070-642:  MCITP Achieved

Tuesday was my wife's birthday.  Between the 070-646 and 083-640 studies, and my error in going through much of the 070-642 book prior, I took the time to sleep well Monday night and didn't rush the morning of my wife's birthday.  I glanced through the topics of the 070-642 materials at lunch before heading out in the afternoon.  Because it was my wife's birthday, I didn't get overly anal with reviewing my answers, and left things to my original judgment.  I think I only looked at four (4) inter-related questions in my review, and got out very early.

It took way too long after I hit Exit Review for the score to come up.  As a joke, I told myself, "well I got a 832" and then said, "either that, or I didn't pass with a 632."  Right after those brief few seconds, the score came up, "833/1000" (700 passing).  I guess I knew what I knew and it was good to go with my first instincts.  The score sheet's graphical breakdown also explained what I knew, and what frustrated me.  DNS and DHCP always frustrate me on Microsoft exams because they use non-standard terms (let alone change them every 2 version) and the questions are always poorly worded in general in ways you'd never proceede in a production network.

E.g., everyone knows that you need to 1) define at least one DHCP scope and 2) active a DHCP server in Active Directory before starting DHCP on a server.  What Microsoft loves to do is throw you questions that test if you've ever been stupid before with that (I can't talk about the exam or questions, but read into what type of questions could be used to test if you've been stupid before).  Methodical people like myself always follow practices, and never do this.  It's one thing to test troubleshooting.  It's another to test people on how Microsoft products break when you don't follow practices.  Again, it's not troubleshooting in my opinion, but alas.

Refocusing on my RHCA, then possibly CEH/CISSP, then MCITP: Enterprise Administrator

I'm still waiting on my official Microsoft transcript to show the MCITP: Server Administrator.  It's clear it takes at least 3 days (if not a full week) for anything to come through:
  • +1 day for the Prometric transmit from Microsoft (and you know when they do it because the MCP site goes down with "Server Error" as they remove access from the public network -- yeah, still like that after 5+ years ;)
  • +1 day for the Microsoft system to take the 3 digit exam suffix and apply the proper MCTS
  • +X day for the Microsoft system to put 640, 642 and 646 together as the MCITP title
When I passed EX436, Red Hat immediately issued the Certified Datacenter Specialist (RHCDS) at the same time as the "Clustering and Storage" certificate of expertise, let alone that was just a few hours after the exam.  It went into the on-line validation.  I guess that's you get for controlling the system, end-to-end, although it was already late on a Friday on Arizona time and the grading was done by someone even later on EDT.  Red Hat states it can take up to 3 weeks for results, but 3 hours is not uncommon at all (and 3 days in the worst case).

I'm sure a lot has to do with the fact that both Prometric and Microsoft are using back-ends that are so outta date, and not always under the control of themselves (or even using their products in the case of Microsoft).  One of the advantages of Red Hat "eating its own dogfood" -- both internally as well as avoiding partners who do not "eat open source as their dogfood," is that things get done and done right.  I'm not saying everything that every Red Hat partner does is always open source, and even Red Hat has acquired firms that produce software for Windows in addition to Linux and/or MacOS X that requires use of proprietary components at times in development and test.  I'm just saying that the whole 083 "performance-based experience" has left me with a greater appreciation for how Red Hat tests, which was also noted by a former Novell-SuSE employee as well.

For now, I'm refocusing on studying for my EX442 retake.  Out of the six (6) Red Hat exams I've taken -- RH301 (cold turkey, exam-only, back in 2003), RH301 (again for re-cert in 2007), EX401, EX432, EX436 and EX442, I failed the EX442 on the first shot with a 66.7%, with 70% passing.  I'm not looking to retaking it, but the next time it's coming to the DC area (which may be only one of two for the rest of the year) is July 24th.  So I have three (3) weeks of study.  The only remaining RHCA exam after that (assuming I pass the EX442 on the second try, let alone 2 years after I took the course), is EX333 which does not return to DC until October.

UPDATE:  The availability of EX333 eluded me until I finally just flew to California in 2011-April.  I passed it and now, at long last, achieved my RHCA.  There are over 90,000 RHCEs, but just barely 300 RHCAs when I recently inquired.


After the RHCA, I plan on taking the Ceritifed Ethical Hacker (CEH) and finally getting around to putting in the paperwork and sitting the CISSP.  I've put the CISSP off way too long.  I know the CEH is newer, but I heard it is not difficult to pass with study.  I know there is the crash course for it, but I've found plenty of materials, including several material free via my IEEE CS membership access to Element K.  Same for the CISSP, although I've had several colleagues who were studying for the CISSP, took it, and then told me I'd have no difficulty passing it.

Post-American football season and into the winter, I suspect I'll get around to retaking the 070-620, and then taking the 070-643 and 647 for the full MCITP "Enterprise Administrator."  After my "experience" with 083-640, I'm desiring to sit the 070-643 before they turn it into an 083-643 "experience" as well.  I'm all for performance-based testing, but I want a local system (or at least a command line into a remote, not a pig of a GUI over framebuffer and that latency).  I'm also wondering if I should just wait for Windows 7 and the "Client" version of that exam, instead of bothering with Vista.

We'll see.  Until then, there's another acronym to put on my resume, and the continued explaination that the MCITP replaces the MCSA/MCSE.  As always, my resume reads various statements such as ...

"Certifications do not represent and should never replace any requirement for experience.
In fact, certifications may be found attached to applications who have no experience whatsoever.
Experience must always be evaluated independently of certifications, and in the context of the available position.
Vendor agnostic experience, based on understanding of technologies, are always preferable to product-tied certifications (or even vendor-agnostic certifications for that matter).
Do not filter candidates based on certification alone or your hiring managers and technical leads may miss the most qualified candidates"
About this Entry
May. 23rd, 2009 @ 01:48 am The $200 netbook, 2 days of power, Internet, Office and no Windows ...
Current Location: Oviedo, FL
Current Mood: anxiousanxious

I have not yet bought a "netbook" for myself.  I bought one for my wife, a Pink Acer One fully loaded (Atom N270, 1GiB RAM, 160GB disk, 6 cell battery, 0.3Mpix webcam and Windows XP) when it hit $260 at Amazon earlier this year.  But I have not purchased one for myself.  I've looked at a few 9-12" Intel Atom based Netbooks, but came up disappointed.  I have even considered  a 12" AMD Neo (under-clocked Athlon 64) with 4GiB RAM and AMD-ATI 700 series chipset from HP that smacks the Intel Atom netbooks silly, but really made me wonder what I was buying for $600 at that sizeable 12".  I almost pulled the trigger on a small 7" ViA C7 unit for $200, but the only half-gig of memory just killed it for me.

Memory is really the big thing for myself, which is why I even looked at the Neo with 4GiB (but at 12"?! Com'mon!).  Memory is always the limitation in performance.  Out-of-order (Neo and traditional CPUs) versus in-line (Atom, C7, etc...) isn't a huge deal, and I can deal with waiting on some launches or calculations.  No need for a hard drive (that's a Windows requirement).  I just want a 8-9" netbook of 1024x600 resolution, something that just leverages commodity DDR2 RAM where I can put in a 2GiB SO-DIMM for $20.  And give me two (2) full days of six (6) hours of computing each -- basically plug my Netbook in every other day like I do my cell phone.

As far as cell phones, I stopped that "rat race" years ago.  I just started carrying unlocked Blackberry units of 1-2 generations back.  Whatever cost under $100/unit, unlocked (even if refurbished, I could get a replacement plan for $10-20 for 2 years).  First that was the 7230, then the 7290 by 2007 and now it will get me a 8700.  It's all I need.  EDGE speed over GSM is more than efficient for getting e-mail pushed to my unit, occasionally looking up something quickly in its browser and doing what it's supposed to let me do -- talk to people.

But for most people, myself included, there is just too much of a jump down to a cell phone and too much of a jump up to a notebook.

The Intel Atom Netbook World, and Why It's Still Windows

Linux Netbooks are not selling.  Let's get real here.  First off, they are still x86 (PC) based.  That's Windows' world, at least Windows XP (Vista and the forthcoming Windows 7 are slouches on Atom).  Secondly, the EEPROM versions are no cheaper -- despite 8GiB of MLC NAND EEPROM being much cheaper than a 1.8" 4200rpm 120-160GB disk -- which removes another staple advantage of Linux.  Third, Microsoft is subsidizing the Netbook vendors like they do any Tier-1 PC OEM, so that removes yet another area of Linux intrusion.

Heck, I'd be happy to just have a POSIX shell command prompt and OpenOffice.org on anything, it doesn't have to be Linux.

The battery life of the Netbook really just sells what it is, it's a notebook, only ultra-light and with longer battery life.  My wife gets up to 6 hours with her oversized battery that sticks out the back of her Acer.  The Atom N270 and its video/peripheral chipset with physical disk eat up 20-25W, as little as 15W when idle.  That's too much to qualify it as anything else but a PC notebook.  To bring a non-Windows world to people, we have to go cheaper and lower, with many advantages.

Heck, PC sales are booming again in volume numbers -- although not fiscal numbers -- thanx to the new age of the inexpensive Netbook of $400-600.

Atom and ARM, the Gap Between Notebook and Handheld, and the Partial Netbook Bridge

Intel abandoned its ARM legacy, a legacy that it acquired from Digital which was sold off piecemeal by Palmer in the late '90s (who was probably the smartest man for doing such in the age of blind .COM belief), back in 2006  Intel put its faith and future into Atom, an in-order, 2-issue x86 redesign that it now leverages for an unified embedded/portable solution.  Right now Intel has that hold on both the Netbooks as well as the home tweakers'n geekers that believe Atom is a "Greener Solution" (in the case of Mini-ITX solutions, that's not the case as I mentioned in my most recent entry).

Unfortunately for Intel, Atom will never be an option for handheld devices.  In a world where ARM and a few other cores dominate, Atom can't even offer itself when cell phone battery life would drop to under one hour talk time.  That's the world Intel has written off.  So far, it hasn't hurt them.  It doesn't want to deal with ARM anymore, probably for the same reason why IBM finally ended offering a generic Power processor for Apple and others.  Then again, IBM didn't drop Power, they just moved over to a higher volume, consoles (e.g., Playstation 3, XBox 360, etc...).

ARM is the player in the handheld mobile market.  It has been for some time.  It will be for some time.  More and more mobile phones are pushing the fine into portable computing.  Unfortunately they have small screens, limited memory and limited keyboards.  Many vendors have tried to bridge the gap between notebook and handheld over the years, but most have failed.  Much less the costs has been prohibitive, the computing limited and the asumed Windows compatibility requirements.

The costs have no come down with the Atom-based netbook.  As 3.5" disk has moved to 2.5" and made it commodity in servers and, increasingly, desktops, the ultra-light is doing the same from the 2.5" disk to 1.8" disk now in netbooks, let alone solid state disk (SSD) options.  The 8-10" LCD of 800x600 to 1280x768, typically 1024x600 today as 16:10 aspect has taken over, has also dropped.  Add in the other, miniture features and details, and the bridge in costs to lightweight, but the cost were partially bridged thanx to 2007-2008 netbook developments.

Consumer Desires and Windows Biggest Problems, including Mobile and 7

Consumers are ready for the next step in portable computing.  They have been for a long time.  These days people sign 2 year contracts so they can save $200 on that new, $500 cell phone that has everything.  I guess I'm abnormal in that I'm not, and I'm happy to have "upgraded" to an unlocked GSM Blackberry 8700 for $100 in the last year.  But for most consumers, it's not always about cost, it's about portability.  It's about not having to be plugged in for power.  It's about being truly mobile.

Consumers want, no expect, the following in a portable computing device: 
  • At least two (2) days of battery life with regular use -- so figure two (2) days of a good six (6) hours of "computing time"
  • The ability to get a full browser, with full features, and popular applications, including graphics editing and web
  • The ability to fully edit common office documents, with full features, and view PDFs with good DPI (dots per inch)
  • Enough memory to run both the browser and an office program or two  without shutting any of them down
There it is, pretty simple, no?  Get "computing time" with their portable computing device like they do their cell phone.  Heck, be like the cell phone vendors and make a killer profit on the sale of a 2nd (or even 3rd) battery.  As long as it lasts for at least two (2) days with heavy usage, they're all for it!

The next two are where Windows runs into trouble.  Microsoft has its CE-based kernel, powering PocketPC / Windows Mobile -- even on processors like the ARM -- but you're not going to find MS Office running on it, just some "applets."  Same with the Internet Explorer and browser.  This only gets worse with the fact that none of these phones are PC (x86) processors, so most Win32/x86 components of the full versions will never be ported for that very reason, and documents and sites will always "just break" anyway.  Microsoft still has that portability, PC-only compatibility nightmare and they've never fixed it -- even with MacOS X (and why they gave up with Internet Explorer on Mac).

Even features in MS Office 2008 for Mac have actually regressed from the MS Office 2004 for Mac -- and caused more compatibility issues with MS Office 2007 users on Windows than MS Office 2003 Windows users had with their Mac friends.  Given Microsoft's fully admission that they aren't adopting their own ISO standard Office Open XML (OOXML) until version 14 (MS Office 2007/2008 is version 12) at the very earliest, it means document compatibility will continue to be an on-going nightmare between versions and non-Windows systems.  MS Office 2007/2008 (version 12) also eats up double the memory if you launch both new version 12 documents and older, version 10-11 (XP/X-2003/2004) documents which launch the "compatibility mode."  Compared to OpenOffice.org, it's a pig -- especially when you consider OpenOffice.org 3.1 actually reduces requirements, has a full, native Mac mode (and a Solver and basic VBA too -- things Microsoft removed in 2008), etc...

And let's talk Solid State Devices (SSD).  I personally cannot wait until Windows-based netbooks with SSDs start crapping out within 1-2 years of sale, outside their basic, 1-year warranties.  Understand the NT 5.1 kernel of Windows XP doesn't know jack about how to buffer or at least overlay (like Embedded XP) writes to the commodity NAND EEPROM aka "flash" in these systems, and NT 5.1 does not offer a "read-only" filesystem (let alone boot) mode.  Wear-leveling does not work at the device level for Windows, it wears out rather quickly when the OS is atop of it.  The NT 6.0 kernel in Vista doesn't solve the problem at all.  Vista only offers a "boost" mode whereby if you have an EEPROM device, it will relocate read-only binaries and files to it, to improve performance over a hard drive -- but it too offers no read-only or overlay mode.  But aren't SSDs "faster"?  No, not at writes, not in the least bit, at least not in commodity versions (long story, read my prior entry and the LinkedIn post).

Oh, but Microsoft Windows 7 will solve these problems, right?  Yeah, right!  Windows Vista won't even run on Atom and any of these processors because of the sheer requirements, and not even a future ARM would be a likely candidate.  Windows Vista, the NT 6 kernel and all of its core libraries, are a pig and not changing in Windows 7 which is -- tada -- a variant of the NT 6.1 kernel in their Windows Server 2008 product (the "improved over Vista" version).  I.e., if you haven't noticed, all netbooks come with Windows XP even though Microsoft isn't supporting it very shortly (at least not without a serious, 7 or 8+ figure licensing agreement from an PC OEM,  Enterprise or on a per-agency Government customer basis).  So Windows 7 is not going to solve the problem, but just re-enforce it.

Maybe 8-core Atoms and 8GiB of memory will hit for $5/each soon, and make Windows 7 an option with a browser and couple of office programs running.  But in reality, don't hold your breath, especially since battery life will be laughable in such a Windows 7 solution.  Consumers need enough RAM to run a basic OS, a featured browser and a couple of office programs, and the OS and office suite can't be pigs (and try to balance things with "compatibility modes" that double or even triple memory and processor usage).

It's Never Been About the Year of the Linux Desktop, but the Killer App on Linux

In 1994, Linux took off because the web hit and it was quickly the open source platform for the open source web server.  People could care less what was running their web site, they only needed the content to be served from it.  In the early 2000s, the same happened with the Digital Video Recorder (DVR).  People didn't even know their TiVO and, later, countless off-shoots were running Linux.  It had a menu and it worked as designed.  Google and their endless apps have made the Internet viable for so many things, as people leverage Google Earth and its APIs, utilize Picasso and other tools that "do the basic things they need," and read their e-mail.

Linux will never, ever be about the "marketed desktop" or "breakage of the distribution lock" that Microsoft has.  As such, I find the phrase "the Year of the Linux Desktop" to not only be only a distraction, but a total misunderstanding of the realities of the computing world.  People use applications, not operating systems.  People know marketing, not technologies.  Linux is not an application and Linux has no marketing -- at least those few successful and profitable Linux and open source companies do not spend money on marketing or trying to break those distribution locks that Microsoft has (my employer being an ideal example).

But Linux can very much be about the next killer app consumer must have, even if they never see Linux.  Sure, they know they aren't running Windows, but that's not always a bad thing.  DVRs "just work" because they don't run Windows, whereas Windows media PCs break down and crash regularly.  ATMs and various information systems (e.g., those at the airport) regularly do that run Windows, and anyone technical notes it.  Heck, the 2008 Olympics couldn't get off the ground without a huge notification to everyone that a Windows system had crashed and the London Stock Exchange (LSE) is beating itself over its head (or at least its Microsoft and its core software development partners for the integration/portability issues and resulting downtime now over 2 years into the effort) for not just following everyone else in going Linux and leveraging existing, POSIX/UNIX code atop of it with minimal effort.

The Era of the Linux Netbook, Multi-day Portable Computing for Most Everyone

That leaves Linux, or at least open source and/or UNIX-based computing that is already running on ARM.  This not only includes nVidia who is building their commodity graphics around ARM in a single package (something Intel has still not done, but nVidia's lower power cores smack Intel's aged or even newer PowerVRs silly -- 6x over), but even Apple.  Remember, Apple has their own UNIX-based OS (FreeBSD compatible) at the heart of Mac as well as their Safari browser built on KHTML (open source from the popular Linux KDE desktop).  But will consumers accept it?  Here's the reasons why they might ...
  • The 2007-2008 era of the notebook evolution as brought the support components -- LCDs and other components -- and will hit at $200, below Intel Atom netbooks (and even ViA C7 it seems)
  • The single package ARM cores + graphics and peripherals will make multi-day computing (2 x 6 hours) on a single charge possible, "charge it when you also charge your phone" and "carry a spare battery or two for a full business week"
  • Performance will match that of Intel Atom, at 1/4th (or lower) power, even after factoring in graphics (especially since Intel continues to ignore it) and peripherals
  • Linux runs extremely well out of commodity EEPROM SSD, because it can read-only boot and use filesystems such, and buffer/overlay writes (especially if the vendor leverages JFFS2 and its write leveling)
  • The Linux distros to support ARM (and even SSD w/JFFS2) already exist -- from Debian to Fedora, which offer native ports and cross-compiling from full PCs in their software repositories -- let alone Google's gOS and other developments, before even considering Apple's possible product
  • The SDHC, in addition to USB devices, is quickly becoming the defacto standard for portable storage, capable of over 30MBps reads and over 10MBps writes in new Class 10 devices (actually over 20MBps, but fliesystem operations managing the files over the 4-bit interfaces cut that to about half), which is now in all devices (even many cell phones) for next-to-nothing, but that performance is really no slower than 1.8" 4200rpm hard drives anyway
  • Mozilla developments like Firefox (or Apple's Safari browser based on KTML) and the leaner footprint of OpenOffice.org over Office 2007 (while often being more compatible, especially with Office 2000/XP/2003) will really reduce Windows compatibility considerations given the benefits of the platform's mobility/longevity/capability -- again, not just an ultra-light notebook, but a true netbook
The only concern I have is the memory that I expect to see in these ARM-based netbooks.  A lot of vendors will try to get the memory size down to save money, shipping only 512MiB (0.5GiB) I'll assume, just like most of the Atom-based Netbooks running Linux.  While this may be the "default option," I plead with these vendors to use a DDR2 SO-DIMM, even if a very simple pin-out and bracket that won't survive more than a change or two.  The worst thing they could do is not provide a 2GiB option for customer.

Because memory size is performance, period.  A processor can be slower, disk writes can be slow or slow writes to flash, but when memory is low, there is paging or, gasp, out-of-memory (OOM) conditions.  Memory means performance in all cases, because no memory available means 100-1,000x performance loss.  DDR2 memory is so commodity, and available in direct pin-out in ARM deisgns these days, that it's a no-brainer to offer it, and leverage the availability of up to 4GiB per channel sizing.  That's the biggest problem with the current Atoms, and their continued use of DDR (and that 1GiB sizing limitation).

Again, that is my only fear in these new ARM-based netbooks.  Memory will be neglected for cost, and the lack of a socket for the same.  Heck, if you really want to save on costs, manufacture and ship the system with no memory to integrators/resllers, but with an easily accessible SO-DIMM socket -- force integrators/resellers to add it, or the consumer, and they cover it.  Lack of memory in the GiB sizes will definitely kill the ARM-based netbook running Linux as a viable option, because 2GiB can make Firefox and OpenOffice.org an usable solution for many that would have never considered Linux.  Because of what the ARM solution itself will be capable of in features, as a full desktop, that Atom cannot, let alone anything running Windows.

Related Articles ...
About this Entry
May. 21st, 2009 @ 02:00 am Hardware Blurbs: Ion+Atom kicks Intel ITX, Desktop 2.5" Backplanes and Stop with the SSD Non-sense!
Current Location: Oviedo, FL
Current Mood: restlessrestless
Tags: , , , ,

It's been a bit since my last blog post, but here's a few things I've wanted to note:
  • Ion+Atom kicks Intel ITX
  • Desktop 2.5" Backplanes, and
  • Stop with the SSD Non-sense!
nVidia's Ion Finally Boosts Atom at Only 60% the Power

As a degreed EE, the only thing that gets to me more than the "Going Green" non-sense is "Save the Planet." First off, it's "Sustain the Planet" (if I have to explain why, then I've already lost you). But, secondly, it's not "Going Green" if you're just following mass assumptions of the (self-censored) or marketing that prys on such (sorry, had to say it, then censor it). Probably the biggest issue is the joke of the Intel Atom in the popular ITX small form-factors (SFF) when Intel continues to pair the 4-8W Atom processor with a 22W+3W power sucking i945GC+ICH chipset, resulting in consumption of 40-45W total. Why does the chipset eat that much? It's because Intel is still recycling the 5+ year-old, 130nm fabbed chipsets on them, that's why.

I was looking forward to some AMD Neo solutions, which are essentially a clock reduced 64-bit core with their 8-11W 700 series chipsets (more for the higher end 790FX) that are fabbed at 55nm by TSMC. Unfortunately the Neos are still 65nm processors, and AMD isn't making the most efficient cores these days. Sure, the single core Neo will whip even the dual-core Atoms, but at a power cost that makes you start to consider just a full, but more energy efficient (power/watt) AMD processor with the same 700 series chipsets. On the portable side, the only Neos seem to be 12" as well, and even though they support 4GiB (see more on that below regarding Atom), one can get a powerful 12" for not much more cost or power consumption.

Thankfully, nVidia's new Ion has become available, making Atom viable on the desktop again (possibly more viable on the portables too). What is Ion? Basically the GeForce 9300 chipset for the Atom. Of course I cannot comment on Linux compatibility of the nVidia chipsets, but outside of the GPU, GPL support should be solid (or so in short order), or at least the PCI IDs supported in short order. The GPU is more of a debate (closed v. GPL drivers). But in any case, as AnandTech found out with the new Zotac board, the power consumption is literally cut to 60% the i945GC+Atom. In other words, expect 25-30W consumption instead of 40-45W with that i945GC heat piper.

Of course, it's still more than the sub-20W you'll get out of a proper Acer One or other portable, that actually has a real, die-shrunk version of the i945GC for portables. But it'll also typically have a slower Atom as well (depending on the model). And the Ion isn't cheap, expect to pay 2x (or more) for the Ion boards with an Atom 330 versus a recycle 130nm i945GC chipset with the same. Now you're back to the age-old, "how many hours do I have to run this to save money?" like the classic, "how many miles do I have to drive to save money?"

Although one thing that has prevented me from buying an Acer One (like my wife has), is the limit of 1GiB pre-installed (1.5GiB if I want to rip it apart), The Ion also offers DDR2 and other support, which means it can support 2GiB easily, if not 4GiB. Intel has its newer Atom chipsets coming out with DDR2 support as well, so there will be another option. But in any case, if you really want to "go green" and mean it, buy an Ion+Atom combination today, not a recycle i945GC power hog with Atom.

As a bonus, you'll actually be able to play lower end 3D titles like World of Warcraft (if that's your thing) on the Atom 330 at just under 20fps -- instead of only 3fps with Intel's uber combo-power/performance sloucher.

Desktop 2.5" Backplanes Reviewed

As many have probably heard me note many times, I stopped buying 3.5" drives in 2006 entirely. I have largely used portables since 2004, and entirely as my "100% desktop" with 17" "desktop replacement" portables -- namely a HP dv9000z (2006) and Gateway P-7811FX (2008). Since that time, I've always updated my portable drives in pairs (since 17" portables can fit 2 internally), recycling the old units to my server. My server now has 160GB, 250GB and 320GB drives, utilizing twelve (12) hot-swap bays.

As I've stated elsewhere, my preference for 2.5" is as follows ...
  • It is the standard for enterprise drives -- i.e., all 10-15,000rpm drives are 2.5" platters, and have been for years
  • Commodity 5400-7200rpm 2.5" drives have MTBF/failure rates based on 10x+ the shock and other operational tolerances than commodity 7200rpm 3.5" drives (a major factor for removable/external drives, even internal in many aspects)
  • I've now had three (3) 3.5" drives fail within two (2) years whereas I've yet to have a single 2.5" drive fail in three (3) years
  • Four (4) 2.5" drives (at least at 5V @ 500mA, 2.5W, or less at full seek) consume less power and deliver more aggregate throughput and independent seek than one (1) 3.5" drive (typically consuming well over 12W to even 15W+ at full seek, sometimes as high as 7W even idle) -- something to consider for, again, even the "Going Green" crowd (sigh)
There are several options for hot-swapping 2.5" drives, assuming you have a controller that supports such, or at least have a procedure to always power-down before doing so. There are the (2) 2.5" in 3.5"x1" bay "trayless" options from several vendors now. Then there are the (4) 2.5" in 5.25"xHH bay "tray" options that I've adopted in my server -- three (3) to be specific, for the twelve (12) bays. The "trayless" option is great for more desktop backup (as I advocate 2.5" removal for most, short-term consumer/desktop backup, at least until NAND SSD drops more in price).

Tom's Hardware has looked at some of the four (4) 2.5" in 5.25"xHH bay "tray" options, which are almost all of the same design -- sans SATA v. SAS multi-port options (with the x4 SAS concentrator cable). I have both and I'm not using any SAS drives, so it makes no difference. The locking is plastic on all of them, so they are clearly not "enterprise quality," but NewEgg carries one brand for $50/unit, so one cannot complain.

The other, nice detail about adopting 2.5" SAS/SATA bays is that the next generation of SSDs over the next couple of years will likely become viable backup (not operational) options for most desktops/servers, and they will utilize 2.5" SAS/SATA form-factors/connectors as well.

Stop with the SSD Non-sense!

Repeat after me ...

"there are different types of solid state devices (SSD), and the commodity SSD available today writes slower than disk, especially over time"

I'm honestly tired of people who assume SSD is Synchronous, Dynamic Random Access Memory (SDRAM) based.  It is not.  Virtually all SSD is Electronically Erasable, Programmable Read Only Memory (EERPOM) aka "Flash" of the NAND gate variety, typically multi level cell (MLC) and not single (SLC).  What does that mean?  The individual cells are very slow at writes and typically wear out with only a few hundred writes per cell.

No, you will not get 100,000 writes out of a cell.  No, you will not see wear-leveling see your SSD last 5 years with any Windows operationg system.  You will not get 100MBps write performance.  Yes, you can get quite fast read performance, and it's great for some operating system aspects (especially Linux).  But no, you're better off with either hard disk or a combination of hard disk and flash on-board for read caching only.  That's what the NT 6 kernel in Vista can do, read cache commonly used binaries, libraries and static (non-changing) files.

And even for Linux, as several of my posts in the Linux Expert section of LinkedIn tried to explain (along with SRAM v. DRAM v. NOR v. NAND, etc...), you have to take care to target select filesystems.  So stop with the SSD non-sense!  Please!  SATA SSD may become viable for backups very soon, as the price continues to drop, but not as the main, secondary storage mechanism in most general purpose OSes for non-expert integrators.

About this Entry
Mar. 29th, 2009 @ 02:58 pm DigitalRiver.COM has removed my Sun GlobalSpecials account entirely ...
Current Location: Oviedo, FL
Current Mood: confusedconfused

For those that haven't read it, here was my prior blog entry on my futility in trying to download my Sun StarOffice 9 binary for Linux after my purchase from Sun GlobalSpecial store c/o DigitalRiver.COM: 
Digital River screws-the-pooch for Sun's StarOffice 9 Linux release

Apparently I wasn't the only one, as many others experienced the same on the Sun forums ...
Corruption with file "release so-9-bin-linux-en-US.sh"

I would like to find one (1) person who downloaded the Linux version of StarOffice 9 between November 2008 and January 2009 that didn't get a corruption.  I'm having the feeling that the Linux binary on DigitalRiver.COM, at least between those dates (I cannot verify other dates), is corrupted, probably due to their upload method.  It also wouldn't surprise me if their Windows version is not corrupted, hence why only us Linux users have an issue (and are far less numerous, especially considering OpenOffice.org 3's inclusion in most Linux distros).

But even after all that, it only gets better.  You see, I've bought every, major version of StarOffice since version 3 back in the mid-'90s.  That includes StarOffice 8 from Sun's GlobalSpecials store site serviced by DigitalRiver.COM.  So my account with my personal e-mail address has been in their system for years.

Today, for the heck of it, I tried to enter my old order number and see if I could retrieve it.  My password didn't work.  Oh well, I figured being that it's been over four (4) months, the access to download the software was removed for my order.  But just for giggles, I tried to reset my password.  My e-mail address is no longer in their system, specifically, "invalid."  Okay, weird.  I used to be able to view my purchases from years ago, for prior StarOffie versions.  I guess no longer?

This is really sad as I've been a StarDivision, now Sun, fanboy for almost a decade and a half.  My documents from old StarOffice 3 still come into OpenOffice.org 3 (StarOffice 9) with few issues.  I cannot say the same for MS Office 4 or 7.0 (95), or even MS Office 8.0 (97) when it comes to MS Office 12 (2007) or even older 11 (2003).  I was even interested in paying for the box set, but it appears that the box set is only offered by Avanquest in the UK, not the US (if anyone finds a link on their USA site, please pass it on).

The only thing that has changed since is that I also ordered some Seagate hard drives. Seagate also uses DigitalRiver.COM, and somehow DigitalRiver.COM can't handle more than one reseller?  That's my only guess.  If so, then they've got even more problems.  In fact, with the Seagate purchase, they increased the cost of each unit of the item by $2, which I got nowhere on.

In any case, at this point, I'm avoiding DigitalRiver.COM and anyone who associates with them.  They obviously have issues with both technical and customer service.  I'm not calling for a boycott or anything, as I'm not into such.  I'm just warning people that I have had repeat issues with them.

Just to review, here are my issues ...
  • Repeatable corruptions in the DigitalRiver.COM download link as (quite invalid) MD5SUMs of match across downloads of even different platforms, networks, browsers, etc... (verified by other people as well)
  • DigitalRiver.COM only offers a download link for the Linux binary (or whatever platform was selected at purchase-time), when the StarOffice license allows installation of any binary on up to five (5) systems (I use only 3, one for Red Hat CSB notebook, one for my personal notebook in Fedora, one for my personal notebook in Windows which I rarely use).  DigitalRiver.COM never answered my requests for a Windows download link
  • My entire, longstanding Sun GlobalSpecials store account has been deleted, or at least no longer known by my personal e-mail address, so I've lost my entire purchase history at the Sun GlobalSpecials store -- this was just the "final straw" (likely serious, technical/customer history issues) as they now have no record of my account at all (including prior purchases)
And my suggestions to Sun ...
  • Make the binaries downloadable under a "trial" link as was done through StarOffice 8, which work for 90 days until a valid license key is entered (enforce the key with an Internet lookup, I'd accept that)
  • Ensure customer service is adequate with DigitalRiver.COM, or consider another reseller, as your individual sales do matter as well (even if not as fruitful as volume sales), especially when it comes to your own, customer records (mine seem to have been removed)
  • Raise the price back to $70-80 or even $100 if it will cover these small, customer service details, as I do understand it takes money to do such (even remove the customer support incidents included, I don't need them, I just want the software)
At this point, I'll have to stick with OpenOffice.org 3.  I wish that wasn't the case.  I am willing to pay $100 for StarOffice 9.  I understand that is hardly a money maker, but it does go a little bit towards servicing purchases and related customer support.  Especially since you do get three (3) support incidents with the current, small purchase price.

BTW, I will still continue to appreciate Sun's purchase of StarDivision, LGPL'ing of the codebase and employment of much of the OpenOffice.org development staff.  I'm hesitant to call American Express to dispute the charge because of this, but I must remember it is DigitalRiver.COM that I have the dispute (and charge) with.

If any Sun employee or OpenOffice.org developer has the Linux and Windows binaries for StarOffice 9, I will personally send them a check for $100, with a Memo that $34.95 is to be given to Sun for a StarOffice license.  I will provide my license number provided by DigitalRiver.COM, which I will use for registration.  I believe in paying for music, software, etc..., never pirate and even paid for MS Word 2007 (largely as a DOCX verifier for clients that are very anal on formatting), and not just the "Student and Teacher" or any Academic edition (the full, commercial version).
About this Entry