quuxpad

The Great Zero-Fill Challenge

Sun, 07 Sep 2008 00:55:47 -0700

The Great Zero-Fill Challenge: Do you need tools like DBAN? These guys say no.

Windows Performance Tools

Wed, 03 Sep 2008 21:15:00 -0700

Remember Bootvis? That’s it, on the right. Microsoft made it to help OEMs setup XP systems that would boot faster. But more than a few techies got hold of it and used it for similar purposes. It provided a very handy view of which processes took the most time during the boot process, allowing a savvy user to tweak things. MS stopped providing Bootvis on their public website, saying that it was of no use to the average end user

I had thought there was no Vista/W2008 equivalent to Bootvis, but it turns out I was wrong. Microsoft makes the Windows Performance Tools (WPT) available for download. WPT includes the On/Off Transition Performance Analyzer so you can track performance data during startup, shutdown, standby/resume, or hibernate/resume. I haven’t had a chance to play with the On/Off perf tool yet, but WPT itself is nice.

Unlike Perfmon, WPT is not live performance information. First you run the xperf tool for awhile, capturing data. When you have completed your data capture, you run the graphical analysis tool.

Above, WPT is showing CPU usage by process, disk reads and writes by priority, and finally disk utilization by process. The first pane is not new - perfmon provides this data. The middle pane is an interesting breakdown of disk reads and writes by priority. And the third pane is really interesting: disk utilization graphed per process and by priority (Perfmon will do overall system I/O per process, but that includes file, network, and other device I/O). The presentation is nice - you can hover the cursor over any line to see which process was using that much of the resource at any one time.

Now here’s a fun bit:

The process lifetimes in the top pane are something you won’t get in Perfmon, and they are quite handy because they sync to the other graphs. The bottom pane here is showing hard faults (agan, available via Perfmon). But the nice bit here is, you can show any two panes next to one another, so you can see the effect of starting a process on, say, CPU or disk utilization. You can also overlay any one pane on top of another. That’s handy.

WPT has a few other tricks up its sleeve, which I’ll let you discover for yourself. Honestly, there’s not a lot of new data here; it’s mainly a different view of the Perfmon counters we already know and love. The ability to track this data during a startup/shutdown is a new and appreciated capability though. And I am excited to finally be able to see disk utilization on a per-process basis.

I have some ideas on performance visualization that I hope to write about in the coming days. Meanwhile, I thank Michael Fortin, who mentioned WPT in a fascinating post on the Engineering Windows 7 blog. I’m really impressed with the quality of information Steve Sinofsky (who heads up the Windows 7 project) is personally exposing in this blog, and I hope to see more of it!

Kill-a-watt

Tue, 02 Sep 2008 17:29:00 -0700

NewEgg is currently selling this handy little device for 18 bucks, plus $7 in shipping. And it’s a great price, but wait, there’s more! Now you can get it for the low, low price of $14 (plus shipping), if you use the promo code EMCAKACAF at the checkout stand. For $21 you can have your very own Kill-a-Watt.

I use my K-A-W when I am setting up a new server. I plug the server into the K-A-W and write down the starting amperage (computers draw the most current at startup, when they power fans at full speed, spin up drives, etc). Then I wait till I hear the fans drop down to their normal, lower speed operation, and measure again - this will be typical of the system’s normal draw. I keep all that data in a spreadsheet, usually sorted by which breaker and/or PDU each server is plugged into. It then becomes easy to see when I am nearing an overload condition.

A problem I have seen in several server rooms is, things work fine until a power outage occurs. Once power is restored, it can be difficult to power the room up again, as all those high starting loads trip the breakers. Similarly, servers draw more power the hotter they get, so if you have any HVAC problems, you can start seeing breakers trip out.

With a $21 outlay and a few more minutes during your server commissioning procedure, you can completely avoid these kinds of issues. There are many other fun and useful things you can do with the K-A-W, like precalculating the power cost per year of a server, or figuring out how overloaded your UPS is. The K-A-W has a readout mode which shows that pesky ‘VA’ number that UPS makers are so enamored of.

It’s a convenient little truth-teller.

http://www.google.com/googlebooks/chrome/index.html#

Tue, 02 Sep 2008 06:55:33 -0700

http://www.google.com/googlebooks/chrome/index.html#

Microsoft DHCP Team Blog

Mon, 01 Sep 2008 23:33:00 -0700

Microsoft DHCP Team Blog: They’re looking for feature suggestions.

IE8 beta 2

Thu, 28 Aug 2008 00:58:00 -0700

Short and sweet: I have found no showstoppers, and several things to really like, in IE8 beta2 (running on my Vista system). If your web browser isn’t mission critical, go ahead and give IE8b2 a try. It won’t bite.

IE8b1 did bite. It rendered many pages terribly; required a close-and-reopen cycle to switch to compatibility mode, and didn’t understand anchor tags. It stayed on my system less than an hour before I reverted back to IE7. Which is a nice thing to know, by the way: if you have problems with the beta, simply uninstall it (the normal way: Control Panel, Programs and Features), and your system reverts back to the IE7 you had before.

IE8 by default renders every new page in standards mode. That makes web devs happy since they no longer have to apply IE-specific hacks to make their pages look nice, but sometimes can throw you for a loop, because pages with IE-specific hacks can actually look bad in standards mode. It’s not a problem though; just click the compatibility button and IE8 will re-render the page in what I’m calling “IE hacks mode”. It will also remember that setting for the next time you visit that page. Nice.

The biggest improvement in IE8 is speed. I have not done objective measurements but to me, pages seem to load a lot faster in IE8, and that is indeed a welcome improvement. Other improvements:

CTRL-F (Find) now does not create a dialog box that blocks your view of the page. Instead it opens a toolbar at the top of the page.
Tabs are now isolated from one another. One hanging web page doesn’t kill all your other sessions.
When you close IE8, there is no more ‘remember tabs for next session’ option. Instead, next time you open IE8, you have the option to ‘reopen last browsing session.’
IE8 closes a lot faster than IE7 did. IE7 used to sort of sit on the second-to-last tab it was closing for a long time (a minute or more) on my system. IE8 closes in about 2 seconds.
Tabs are colorcoded and grouped. So all tabs which were opened from one page, share the same color as that page. Handy.
The address bar, once you start typing in it, is a lot smarter. I did note that sometimes I could not tab to the top choice in it, though! Had to actually point ‘n’ click with the mouse. Hope they fix that.

These are just the features which stood out for me as I played with it today. They seem like little things - but for me, they’ve already made web browsing much more pleasant. There are more new things; Ed Bott talks about them, and so does the IE8 team. The release notes (for Vista) are here. This version is said to be feature complete: from here to release, the IE8 team will be working on bug and performance fixes as well as documentation writing.

IE8 is definitely going to save me some time and frustration throughout the working day. It’s a keeper!

Vista and multiple displays: annoyance

Thu, 21 Aug 2008 04:40:00 -0700

This is bugging me. Recently I added a spiffy new monitor to my desk. My primary work system is a ThinkPad T42, which lives in a docking station when it’s here on the desk. I position the laptop so that its screen is beneath the new screen, so I need Vista to understand the relative layout of the two screens.

I can do that:

… but what I find is that every time the screensaver is invoked, when I log back in to the system, I see the screens like this for about 2 seconds. Then the screen flashes black, and my layout returns to:

This also happens after every sleep, hibernate, user switch, or reboot event. Essentially, every time I must login to the system with username and password, Vista goes out of its way to return to this setting. So I must manually switch it back to the preferred layout. I have discovered that this issue does not plague the builtin Administrator user. But it does afflict all other users, includng members of the Administrators group.

Thanks to Michael at the Vista Forums, I found a suboptimal solution. In the Start Menu, type Scheduled Tasks, wait for it to appear, then right-click it and choose Run as Administrator. The new (and much more complex!) Task Scheduler interface appears.

So now I don’t have to reset monitor settings after every login, but I still do have to reset them each time I re-dock the system, otherwise my second monitor stays blank. It’s progress … but it’s not exactly a full solution.

If you know a better solution, speak up!

Side note. Hey, Microsoft - you need to put more work into the new Task Scheduler. It’s too busy. Also you folks need to rethink some of those default tasks you’ve put in there. The ones with GUIDs for names? Come on, guys; you know better. And the ones with “Custom Handler” for an action? Err … right. What does that do, and how do I find out?

About the users

Tue, 19 Aug 2008 14:03:00 -0700

If you’re anywhere in IT, I think you’ll profit by reading Joel Spolsky’s 9-chapter User Interface Design for Programmers. It’s available free on the web - in fact that link opens the whole thing as one big web page. If you’re not a slow reader, you should be able to get through the whole thing in under two hours. You’ll learn some surprising things about how users think. Some are funny at first; but like many of the funniest things, they are also inescapably true. And sometimes, the joke’s on you, Mr. IT Guru!

But why would you want to spend two hours on this if you’re not a programmer doing something related to user interface design, you ask? The answer is, no matter where you are in IT, you serve the users, and you should learn how they think. We’ve all heard the dumb user jokes; we’ve all seen our colleagues grimace at the mere mention of that one especially thick user in Accounting. Most of us have given or at least heard the rant about the idiot users, and how the systems would be so much easier to manage if we could somehow remove the users.

I’ve done it too. But I stand here with a smile on my face and I really mean it when I say: we’re wrong when we think this way. Just flat-out wrong. We need to make serious efforts to change our thinking about this. The occasional comic rant may be ok, but by and large, we need to get it through our IT-biased heads that users have better things to do than become computer experts like us.

Okay, so that’s a bit of a hard line. Joel’s mini-book isn’t as in-your-face about this as I am; it is an easy, funny, and quick read, and it is well worth your time, since you serve the users. Even if you never see or talk to them, you serve them - and learning a bit about what this stuff looks like from their POV will make you better at your job.

Can’t be bothered to read the whole thing? Then I recommend the mere 10 minutes you’ll spend on Chapter 6: Designing for People Who Have Better Things To Do With Their Lives.

This is what they said Vista would be … in 2003. What...

Mon, 18 Aug 2008 22:51:42 -0700

This is what they said Vista would be … in 2003. What happened?

Computer Description

Sat, 16 Aug 2008 06:46:00 -0700

SF-DC-01.

Many organizations name their computers (or at least the servers) like this. LOCATION-PURPOSE-NUMBER or similar. So the above would be a server in San Francisco, primarily tasked with being a domain controller, and having the number 01 just to differentiate it from any other domain controllers in SF. Personally, I am not a fan of name schemes like this, because they suffer a number of naggy little issues:

Systems change over time. If SF-FS-01 (a fileserver) gets an AV signature server service added to it, do we have to rename the server to SF-FSAV-01? Or do we just live with the idea that our names aren’t as descriptive as they used to be?
The computer name field is short. 15 characters is all you get. So you end up with rather cryptic names which can be …
Hard to pronounce. If I’m talking to the guy in our (loud!) SF data center, and asking him to reboot SF-DC-01, but he instead reboots SF-BC-01, avoidable hilarity may ensue!
Names become increasingly more cryptic, or people don’t apply them correctly. Some servers are SF-xx-xx, some are SFO-xx-xx. And what do we do about the servers in Redmond, WA, versus the ones in Redmond, OR? Or the fact that we now have two locations in Chicago?
Name schemes paint us into corners when unexpected situations arise. Do we amend the name scheme and rename all systems to reflect the changes? Or do we live with the fact that we now have two or more different name schemes … and we can’t be sure which is which?

Which brings me to the point of this little article. You can use the Computer Description field to hold all that metadata:

Now from the commandline, with net view, you can do some neat stuff:

Notice how, by piping the command through a findstr, I was able to pull out just the Windows 2003 systems. Obviously there are many other ways I could use this to locate systems of interest. The other thing to notice is that now, names don’t do anything other than make a computer uniquely addressable. We can choose easy-to-pronounce names, and we can let employees name their own systems. The ‘name scheme’ has really moved to become a ‘description scheme’ which is much more flexible because we can change descriptions without changing the address of the computer. Also, since descriptions can be longer, they are easier to read and easier to search. And if our description scheme needs updating later on in life, it will be relatively easy to reset the description field for every system we have. So we can easily escape any corners we might paint ourselves into.

There are a few things to keep in mind. First, the Computer Description field can hold a lot of characters. I gave up counting when I got over 200 chars on a Windows 2003 system (via the dialog shown above), and Microsoft’s documentation of the registry value where this is kept gives no maximum limit. BUT, and this could be a showstopper, KB231312 documents a ‘by design’ 48 character limit, and KB932762 documents a (patchable) issue where descriptions larger than 48 chars could crash the Server service, on which many other services depend. It is not 100% clear whether this patch has been pushed out via the normal Windows Update mechanism, so I recommend staying under 48 chars in your descriptions.

Also, setting the computer description does not necessarily update the description field in Active Directory Users and Computers. I don’t think it would be to hard to write a script which gets all localcomputer descriptions, then updates AD with these values. You could run such a script daily or weekly. I don’t feel like writing that script at the moment, but if anyone asks (via comments), I just might be convinced to do it.

Finally keep in mind that unless you are running WINS, these computer desciptions will not be available for systems outside the subnet you are on. So that might be a good reason to either run WINS, or whip up the script I described above.

But what about non-Windows systems? Well, the various *nix OS’s can all run Samba, but using Samba just to publish system description fields is probably a bit of overkill. You could explore using MOTD, or QOTD to hold that information. Or perhaps TXT or SRV records in DNS.

Wrapping up: I think computer naming schemes usually look good when we’re thinking them up, but without iron-fisted control, often devolve into a sort of ratty and not-always-reliable mess as time goes by. Maybe we should just free ourselves from them entirely!

Troubleshooting pool memory

Thu, 14 Aug 2008 10:14:16 -0700

Troubleshooting pool memory : How to resolve SRV errors.

PC Incubator

Mon, 04 Aug 2008 17:42:43 -0700

PC Incubator: Chooses top-rated components from Newegg to help you build a great PC. Try it!

Observing chkdsk

Sun, 03 Aug 2008 05:36:00 -0700

I recently had occasion to suspect there might be problems on my Vista laptop’s C: drive, so I scheduled a chkdsk (more technical explanation of chkdsk in this KB article). Since chkdsk cannot run on a mounted drive, and obviously one cannot unmount the system drive, the chkdsk was scheduled to run at next boot. But my suspicions were not really urgent, so I didn’t reboot right away.

Two days later, when I did get around to rebooting, the chkdsk surprised me a bit - I forgot I had scheduled it. But that’s just the framing story for two things I noticed during the chkdsk. The first may be a bug. During stage 4 (check for bad sectors, according to KB187941), chkdsk reports 15% the whole time. The reported completion percentage never changes:

84791 / 163056 = 52%. I tried to get a shot of it just as this stage was completing, but at about 140,000 files, the process sped up considerably, and moved on to the next stage before I could get the camera positioned. Anyway, it stayed at 15% right up to the finish of this stage. Other stages did report their percentage more or less accurately (I didn’t do the math). By the way, this stage was very, very slow on my system. It probably took an hour to complete, though I did not time it.

Now here’s the other thing - apparently chkdsk clears the SuperFetch cache. Here’s ATM, a few minutes after the reboot completed:

Note the almost-missing big yellow bar in the graph. Now remember just a few days ago I put up this screenshot of ATM showing a nice fat yellow cache bar:

It’s not a big deal, though. My system has now been running for about 2 more hours, and I have my fat yellow superfetch band again. Still, it’s a nice thing to be aware of.

ReadyBoost Blues

Fri, 01 Aug 2008 02:24:00 -0700

About a week ago, I noted in the winsat article that I was trying to get ReadyBoost to work on my laptop, which runs Vista Ultimate. The promise of ReadyBoost is better performance, mainly when invoking applications. I’ve given up on it, but will record a few notes here for your amusement and my later reference.

Into my aging Thinkpad T42 (type 2378-FVU, now exactly 4 years old), I installed a Delkin ‘eFilm Pro’ DDCFPROCB-AD Cardbus Adapter and a RiData Lighting Series 233x 8GB CF card. I ran the following Winsat tests (as documented by James O’Neill), twice - once formatted as Fat32, and again when formatted as NTFS:

winsat disk –read –ran –ransize 4096 –drive F -v
FAT result: 5.99 MB/s
NTFS result: 8.46 MB/s

winsat disk –write –ran –ransize 524288 –drive F -v
FAT: 5.01 MB/s
NTFS: 5.01 MB/s

These results are not blazing fast, but they do exceed the requirements for ReadyBoost, which are 2.5MB/sec throughput for 4K random reads and 1.75MB/sec throughput for 512K random writes. You can find documentation of those requirements, and a lot more, in Tom Archer’s excellent FAQ. Tom got his answers directly from Matt Ayers, the Microsoft Program Manager who owns the ReadyBoost feature. So this stuff is straight frm the horse’s mouth.

OK. So my hardware meets the performance standard, but Vista will not give me a ReadyBoost tab in the CF drive’s properties. Bummer.

And with a bit more digging, I found the reason:

Well, part of the reason anyway. Device Manager does properly identify both the CF card and the Delkin cardbus adapter:

So, I’m in a quandary. Is this an issue with the CF card, or the Cardbus adapter, or something about my system’s chipset? It is impossible to tell. There just isn’t much information on the web (I have linked all the best stuff in this article, and it is not enough). Plenty of info about ReadyBoost for USB devices, but not much when it comes to this intersection of ReadyBoost, CF, adapter, and motherboard chipset. I won’t keep buying devices in the hopes that some combination of them will work. So, I’m throwing in the towel on this one, until better information is available.

Here are a couple of sites which do document working ReadyBoost hardware, though none had the complete combination of information I needed:

Techcrater list of tested ReadyBoost speeds.
Grant Gibson’s works/doesn’t work list.
Ed Bott’s shorter list of tested USB devices.

Windows perf team blog

Sun, 27 Jul 2008 18:44:40 -0700

Windows perf team blog: Some deep explanations of How Things Work.

Someday, this won’t seem like much.

Sun, 27 Jul 2008 11:23:20 -0700

Someday, this won’t seem like much.

Windows Memory Management

Sat, 26 Jul 2008 12:19:00 -0700

Unless you’ve studied it a bit, you probably have a few wrong impressions about how Windows memory management works, and that may be hurting your understanding of how to get good performance from the OS. So here are a few links that are worth reading. On this topic, there seems to be a lot of just plain wrong advice out there on the web. So here are a few links which avoid those common misconceptions:

Understanding Virtual Memory by Perris Calderon. This will really help form your base understanding of how the backing store works, and why several ‘common sense’ virtual memory tweaks are in fact kind of disastrous.
The above article doesn’t cover improvements to the VMM (virtual memory manager) which are now implemented in Vista/W2003. Mark Russinovich explains the new stuff here.
ArsTechnica has a very clued-in member who goes by the name DriverGuru. From time to time he answers questions about memory and pagefile usage. Those threads can walk you through some realworld examples, and so they help drive the message home.
MSDN has a large section on memory management.
The best reference is a book: Windows Internals 4th Edition, by Mark Russinovich and David Solomon. The 5th Edition is due out soon.
Update: Mr. Russinovich has a new blog article out, which really helps to provide a more depth-ful answer to the question: Dude, Where’s My 4 Gigabytes of RAM?

The above picture is of Tmurgent’s ATM tool, very handy on Vista/W2008 (though it also works on earlier versions of the OS). I added a bit of text to the above screenshot of the tool running on my own system (2GB of RAM installed).

OK, you’re saying, that’s fine and dandy, but you don’t want to deal with a lot of mumbo jumbo. You just want to know if your system is using memory optimally! OK, here’s how to check. Open Perfmon (it’s part of Windows), remove all the default counters it is graphing, and add this one counter: Memory, Page Reads/sec. You’ll get a graph like the one below:

Your goal is to make that line as flat and low as possible while you are using your system normally. That’s the number of times Windows has to fetch data out of the pagefile on disk and stick it back into RAM. You’ll probably never get it to the point where it stays perfectly flat at zero, but you do want it to return to zero much of the time, with only occasional spikes, and those spikes should usually be to 10 or less. Note that in this example, I set the multiplier scale to 10, so that spike to ‘60’ on the graph was really only a value of 6 page reads per second. This is a system that’s responsive and very rarely goes to disk to retreive data that should be in RAM.

If you’re seeing too many page reads, you have two basic choices: 1) buy more RAM, or 2) run less memory intensive processes. Choice #2 can mean a lot of things. It may mean just closing programs you are not using, or it may mean looking around for one or two programs which seem to hog a lot of RAM.

Tweaking your page file isn’t going to do anything to make this graph look better, and in many cases will make it a lot worse. Unless you really really understand this stuff and are in fairly unusual circumstances, tweaking the pagefile or other OS parameters is not going to be a winning strategy. Don’t say I didn’t warn you!

Lazy Linux Admin: 10 tricks

Fri, 25 Jul 2008 17:11:05 -0700

Lazy Linux Admin: 10 tricks

Exchange Admins, read this.

Wed, 23 Jul 2008 11:36:00 -0700

Running Exchange 2003? Exchange 2003 Disaster Recovery Operations Guide

Exchange 2007? Disaster Recovery and perhaps High Availability

Sometime in your Exchange admin career, you’ll be asked to recover lost data. This isn’t because Exchange itself is intrinsically a fragile platform. But if you do not set it up with recovery situations in mind, there’s a great likelihood that when one does come around, you may find yourself filled with frustration and rage.

The key here is to think of these DR procedures as proactive measures; not something you postpone until the day an issue arises. Being proactive will make that day much less likely to occur, and make the (now much less likely) disaster response a much more pleasant event.

I estimate that (depending on your Exchange setup complexity and size) the task of sitting down with this paper and bringing your Exchange to disaster readiness is a 2-3 day task. It’ll save you far more than that in the long run, but you’ll never really know. It’s like tying your shoes: how many broken legs has that saved you from?

EasyVMX

Mon, 21 Jul 2008 10:05:26 -0700

EasyVMX: Use this tool to easily create free VMware Player instances with your own OS build.