f13.net

I can do some halfass perf tuning on AIX but I am rather lost in Windows.  Several of the concepts are similar but the specifics and tools differ, so I would like to get some input from the Brain Trust.

Besides the Resource Monitor, today I found and experimented with the Performance Monitor.  Took me about thirty minutes to figure out how to generate a log and then view it; seems pretty comprehensive as a metric tool so I figure we want to stick with that unless there are some crazy-nuts alternatives.

Here is the problem I am looking at: ATI HD 4800, i7 cpu, 4GB RAM, W7 64-bit, running The Sims 3 (wife PC), and game textures are slow to load.  I'm more than ready to assume TS3 is a crap application, but I want to see what I can do.  But!  I updated the ATI drivers and on the reboot I noticed it seemed to boot slowly.  Specifically I noticed the desktop icons took some time to load.  There is just a single 1TB disk in the system, and I'd like to think W7 could handle that with no trouble but I found on my own rig that once I passed a certain amount of storage that Explorer and many disk operations became slow.  That was XP and the problem persists in a fairly consistent manner now so I expect the issue remains... whatever it is.

How do I find the bottleneck?  In-game texture load times instinctively appear due to LOTS of textures filling gfx memory.  Slow-loading desktop icons seem to point to I/O delays.  Could be that damn ATi piece of shit.

Currently running an aggressive MyDefrag.

How much RAM does your video card have?

Windows 7 is an enigma better left alone until at least SP1 IMO. I'm well aware that it's worked for plenty of people; but if you're having issues now trouble shooting things in Windows 7 has proven to be a giant exercise in futility. I've had fun problems such as...


Windows 7 booting up randomly with a black screen and mouse pointer. For a time, booting under safe mode and doing a roll back solved this issue until booting into safe mode decided to start with the same schtick.

Team Fortress 2 and Left 4 Dead 2 stuttering a lot until I used gamebooster, after that it only stuttered a little bit.

Windows 7 mouse acceleration unworkable. When you're an FPS guy like me, suddenly having a completely different smoothed out mouse with no way to get it back to the way it was in XP is unacceptable.

Windows 7 for some reason not recognizing my Samsung monitor, leading all whites in Adobe Photoshop to appear as cream.

Windows 7 taking a long time to perform menial tasks (such as deleting a text file, or running CC Cleaner)

Applications failing for no clear reason - EX: Malwarebytes deciding to lock up my computer during a scan


Despite near universal acclaim, I just don't get what's so great about this... But I guess after Vista's launch; a pile of dog shit would seem appealing.

Eventually, issues you have may take you from "I can deal with that... I guess" to "Fuck this, I'm going to back to XP". Back to XP myself and the difference is like night and day, honestly... Though I'm not sure if I can contain my despair over losing Microsoft Aero.



If anything, it's probably the video card. It definitely seems like the most dated part out of what he listed; though in theory it all should work fine in 7.

Not sure what your problems were stemming from. Granted, I'm using the final beta still (thanks for a free year, MS), but I've had zero problems with the OS itself. I've had a few isolated random problems involving HL2-engine games, but I've had those same rare problems on my old PC as well.

The card is a single generation old.

Is your hard disk running AHCI mode? The icons and windows itself being slow loading would point away from the graphics card imo.

I assume it is an ICH10/ICH10R controller? Try some hard disk benchmarks with ATTO or Crystal Diskmark and post a few results, along with the HD model.

By default the write buffers are set to a less aggressive mode in windows out of the box. To change them you can do the following:

Start -> Computer
Right mouse click on Computer in left hand pane and select "Manage".
Click on "Device Manager" and expand the "Disk drives" section.
Right mouse click on your hard drive listed and click on "Properties"
Click on the "Policies" tab.

Make sure "Better performance" is selected not quick removal (This appears if you have the drive running in AHCI mode).

*This will disable the write buffers and you may loose data during a power outage*
Check "Enable write caching on the device"
Check "Turn off Windows write-cache buffer flushing on the device"

Your shit sounds fucked up.

How fragged is your drive?  When you installed was it on a clean format or did you just throw w7 onto the drive with some other preexisting OS's (ie is the w7 OS scattered across various crappy locations on the disk)?  May want to go ahead and manually manage your swap file to 1.5 times your memory (6141), to me it looks like w7 might do a slightly better job of pre-allocating swap space than previous versions but somehow I doubt it is as stable as just fixing it to a reasonable amount  (preferably in a single contiguous space).

Also, the default power scheme in w7 is pretty fucked up for a desktop machine.

That could be it. Some motherboards (like mine) have that turned off by default in the BIOS. And of course Windows 7 is so fucked up that it won't boot up anymore if you switch that in your BIOS after you install Windows 7 unless you go through an obscure registry hack before changing the setting (which is what I did).

Could something like that be a potential cause for this guy's problems?


Maybe by the time I build my next machine, Windows 7 will actually work properly for me. I've tried it on four separate occasions and each time it proved to be a giant waste of...time. (The first three times I've tried it: games I played would intermittently stop responding for a split second. Could be during heavy action, no action, any time at all)

I had a ton of problems after installing Win7. I did everything up to and including a clean reinstall of the OS and kept having problems. It turned out I had a bad stick of RAM.  :awesome_for_real: After getting that replaced, I've had no probs with 7 atall.

First reboot after an ATI driver update will generally be a bit longer, I've found.  I haven't played Sims3, but desktop icons and texture issues are generally Random I/O related, from my experience.  AHCI has some issues in various places, but generally the performance should be about the same as running in regular mode.  The defrag you're doing is good, but I would look at a HD upgrade of some sort - faster HD, add a second for RAID0, or go SSD.  Each of those has its own ups (speed) and downs (cost, plus reliability on the RAID0).

Minraven, where did you get that avatar?

The source for all things Peep-y (http://www.peepresearch.org/)

You are going to make me google this? :awesome_for_real:

As for the people who are complaining about Win7, and the ones telling me to buy some new hardware, please read the thread title and top post.  I am interested in tuning this system, if possible, not in replacing a C: that is not the fastest in the first place.  I am probably going to replace this crap-ass ATi card, so let's not bother with those comments either.


Lots.  Being dealt with via MyDefrag in Monthly mode.


It was a "clean" install, in that it was not an upgrade, since it replaced Win7 RC.  I did not format the disk, though.  I think this line is related to your defrag line but let me know if you are getting at something else.


I set it to 4096.  Seems the pagefile is in two fragments but I'll have to review the log again to be sure.  It apparently has 1.2GB of stuff in it after a boot, which I find interesting, but that's why I made a thread about performance tuning. :oh_i_see:


Thanks for the reminder, it was on Balanced so I set it to the Secret one.  The defrag seemed like it had been interrupted; wife claims ignorance.  Have not checked the event log to see if there was a reboot yet.


I don't have this option so I am guessing it is not an issue.


Sounds dangerous.


ASUS P6T Deluxe, so ICH10R.  You also get a star for good advice.  I'll see about finding those tools.  Very well could be the disk design since it is a WD10EADS, meant for cool and quiet, but I'd not have thought that meant it would actually suck.


512MB.

I'll be nice:

http://support.microsoft.com/kb/922976

I appreciate you indulging my laziness in this venture.  That doesn't look to hard but having to do it is annoying.  The advice is to enable AHCI, then?  I glanced at the spec sheet but decided it wasn't going to be terribly useful.  Assuming the advice is to enable it, do I need to go into BIOS to see if it is turned on?

I'm getting the idea that Windows perf tuning is still nonexistant.

Well it is and it isn't, at worst you can loose the data that wasn't written out via your hard drives write buffer. In the grand scheme of things the power outage could do much more damage say it was power spike  :grin:


It is the suck, I have an exact same disk in my WHS and here's my ATTO results (http://www.techpowerup.com/downloads/1137/ATTO_Disk_Benchmark_v2.34.html):

You should at least check your BIOS to see how your SATA ports are configured.

Yes, good idea, it could very well be that the controller defaults to legacy/IDE mode, which is a performance bottleneck for modern HDs.

If the controller was set to legacy mode from start, then W7 was installed without AHCI support and enabling it will most likely result in the bluescreen from Trippy's link. Then you have to reset to IDE in the BIOS and perform the registry hack, reboot and reset to AHCI in the BIOS and you should be good. If the computer was already set to AHCI at some point and successfully booted but resetted, W7 should boot without problems.

Here's a slightly newer ATTO version (http://downloads.guru3d.com/ATTO-Disk-Benchmark-v2.41-download-2343.html) than the one from fuser - although his should work okay too, this thing is so old it doesn't matter much. Also, judging by fusers benchmark results, I would expect results somewhere in the 50-60MB/s read/write range. Anything lower points to a problem.

Since you are defragging monthly. I really don't think it is a fragmentation issue.

If the results for the benchmarks are still bad after enabling AHCI, try installing the intel storage driver from here (http://www.intel.com/support/chipsets/imsm/). I would suggest using version 8.8 even though they say it is unsupported under W7, it gave me better results than 8.9. There's also a preview version of a redesigned intel software floating around somewhere (they renamed it to "Intel Rapid Storage Technology" from "Intel Matrix Storage Manager"), but I'll have to google where I found that.

One of the things I want to learn about Win perf tuning is what numbers can be expected for what metric.  I ran ATTO on my own machine, which is a Frankenstein's-monster in that I generally update it in pieces, but still the marks on my C: are slighty surprising given your comment:
(http://dl.dropbox.com/u/85916/atto-x1-c-1.png)


What I meant was that I was using the "Monthly" script with MyDefrag, which is very aggressive.  I am pretty sure it hasn't been defragged since the most recent OS installation.

Anyway, I'll be looking into whether AHCI is set in BIOS when I get a moment.  I'd like to have some observable metric that tells me I have disk IO latency (and which disk), seems ATTO is the thing to use.

Yeah, I assumed so.  Mostly I don't want my wife to accidentally unplug it and corrupt something important, like the filesystem.  I'll think about it.

My E: is a WD5000AAKS and is almost exactly 50MB/sec in ATTO.  My C: is a WD1500HLFS (don't ask)... so I suppose we all need a new OS disk around here. :oh_i_see:

Those are awesome results. Is that the WD1500HLFS you mention? Cause that drive is supposed to bench that good. A Caviar Black should peak out somewhere around 100-110, a regular or "green" disk somewhere in the 50-70 range. SSDs can give you somewhere around 150-200, 220 if they are fast and fresh out of the box.

For comparisons My Indilinx SSD hovers around 150 read/220 write, Raid 0 with 2 x Caviar Black (WD1001FALS) is in the 120-150 range and Raid 1 (same drives, matrix raid) around 90-100 MB/sec.

Ah, misread that, sorry. Even considering you haven't defragged since OS installation, six months (at most?) of fragmentation shouldn't really kill performance on that level.

Another interesting metric, if you're digging into the matter, is to plot the decrease in transfer rate the further you get away from the center of the drive platter. It could very well be that the stuff that matters in this case (textures from the game) ended up somewhere on the outer rings of the platter, and the transfer speeds decrease towards that. HD Tach (http://www.simplisoftware.com/Public/index.php?request=HdTach) can do that, but it only does it for read access in the free version.

Are these tests affected depending on wether they're the active C drive or just a slave drive? I ask because the 1 TB Seagate drive I got a while back seems to out-perform my brand new 1 TB Caviar Black (not by much tho), but the Caviar is my C: drive and the Seagate is my storage)

Ugh, I can't read.  Of course more MB/sec is better, why did I think it was transfer rate in milliseconds? :uhrr:


That is my C: and the funny thing about it is that is is a 2.5" drive inside a 3.5" cooling tray.  Don't confuse this with the 1TB WD Green in my wife's PC, though.  In my case, my rig has I/O trouble due to... something... which I attribute to having too much storage configured, or too many disks, but considering I haven't used all the plugs I don't think it is hardware.  It just showed up one day as soon as I configured a new 500GB SATA disk.  Another reason for my sudden interest in W7 perf.


I'm waiting to see how the SSDs work over long periods of time.  I am concerned about lifespan.


I am digging.  Lots.  Star for you, I will look into that tool.


Things like this is why I need to figure out how to get the metrics I need.  The Performance Monitor that comes with W7 will tell me a shit-ton of things but I don't really know which ones to pay close attention to.  I am able to see the I/O of various processes but I am certain that I am not seeing a full breakdown.  I also want to know how to know what is paged out.  So much so that I'm about to look for a book on the subject.

A real slave drive, like back in the IDE days with their multiple connector cables, or connected to a secondary SATA connector? I don't really know about the former, but it shouldn't really hurt benchmarks with the latter. At least I don't think it does on my computer, as I have plenty of stuff in secondary SATA ports without noticeable performance losses.

Me too! I bought this one on a cheap deal about 8 months ago, average usage, has been holding up pretty well. I'm seeing no performance decrease and I even managed to do a firmware update without losing data, which totally surprised me. I'm thinking that if it lasts around 3 years, it was well worth the money. Using it as my system drive since W7 and the firmware support the trim command.

It didn't really make a huge difference with games when I tried it, but it is pretty fast with windows and starting up all the stuff from programs on a regular basis.

A decent ssd (ie intel) the lifespan should exceed that of a mechanical drive, the article I read said that the flash used has a fairly predictable lifespan of 10k writes per cell.  The intel firmware takes this into account and spreads uasage around for max performance & longevity.  The person who wrote the article considered his hard drive use to be on the heavy end of the scale and calculated the lifespan of his drive to be in excess of 10 years.

Yeg, if you had pre-existing crap on your drive and it was heavily fragged prior to your w7 install you may be permanently dorked.  Unless you have some magic defrag util that will reorganize your w7 os filees to a large contiguous space at the start of your disk.

If you want your swap to be in 1 peice, put it in its own partition.

Sorry for typos, posting from phone while being subjected to the wife's driving.

Sorry, I wasn't clearer. Its not a real slave drive. Just a second hard drive in the 2nd Sata port on my motherboard. But the real question I have is wether running these benchmark test on the active OS drive (C:) is a real indicator of speed, since all the while Windows (or whatever) must be doing some access to the drive while the test is being run, hence lowering the score. At least that's what I'd expect, but I'm not sure. So that's my question.

It seems I could probably do this with MyDefrag if I took the time to identify the files, but that seems possibly tedious.  Unless it would be fine to match on anything "C:/Windows/*" and sort it to the beginning of the disk.  Hmmmm... /chinstroke

Well, I used to use a MS program that you dumped in /system32 called contig. That would allow you to specify a directory and make sure all the files in there were both defragged and contiguous. For a quick boost, I'd do my /system32, and just /windows in general.

You can find it here among other cool toys : http://technet.microsoft.com/en-us/sysinternals/bb545027.aspx

More specifically, here http://technet.microsoft.com/en-us/sysinternals/bb897428.aspx

I liked MyDefrag, btw. Danged drive was being all noise spastic till I ran it. Now its relatively quiet.

Edit: Just realised that the operative phrase in this discussion is 'at the start of your disk'. Sorry, in that dept, I got nutin.

Edit2: Oh oh, I know. Make your parition as small as humanly possible using Parted Magic (http://partedmagic.com/download.html), then defrag, and then use Parted Magic again to blow open the partition to useable size <grin>

This thread is of interest.

Thanks for pointing out the ATTO benchmark tool, W7 gives me low marks for my main drive for some reason, but this seems to indicate my SATA RAID 0 is working alright.

(http://dl.dropbox.com/u/42220/SATA_RAID_benchmark.png)

However, I don't think I'm running my disks (RAID or other, I have 5 total with 3 in the RAID) in AHCI so I'll play with that and see what I get.

I would worry more about a defragged section for the OS before stressing over where it is on the disk, you would think position on the disk wouldn't matter as much on a cutting edge hard drive.

From what I have gathered, it can make a significant difference with larger drives. The needle hardly has to travel if its all written on the edge of a 1 TB drive, while on a smaller disk, say a 160 GB drive, it won't matter as much. This is anecdotal.

I would still think disk thrashing caused by excessive fragmenting would be a much greater performance inhibitor.  Data density on a 1tb drive is insane and pretty much all of them are using perpendicular recording where this may not the case with a 160gb drive, basically if you have your OS files in a contiguous space on the slowest part of a 1tb drive it is probably going to be faster than the fastest contiguous space available on a 160gb drive.  Naturally there are exceptions such as the raptor/velociraptor drives but even then the top of class 10k rpm drive's only have a minimal speed advantage vs. a decent mid grade 1tb drive.

Well, sure, the first steps are defragging and then contiguous files, but we're going full throttle nerd here.

For a real world scenario, position matters about as much or more than defragmentation nowadays. Defragmentation is a legacy thing from the days when random seek times were killing drives. Although both are more likely to only have a noticeable impact in cases where you are doing reading/writing on really huge files (in the GB and above size) or lots of I/Os in a very short time (which can be a factor depending on how for example a game uses thousands of individual files).

Number-wise, there's often about 20-30% loss in sequential I/O performance on the outer regions of a 1TB drive compared to the center. Whereas you're not losing 30%  performance on unfragmented vs. fragmented drives unless there's really LOTS of fragmentation. HDTach is pretty good for pointing that out.

Do you mean the other way around? The greater the circumference of a disk, the less the needle's spindle/disk axel has to turn, so the farther from the center the faster, goes my logic.

You are correct, I had that wrong. I got confused in my initial interpretation because of thinking that data is being more spread out towards the edges (as block density is the same on all rings, I think?), but the linear velocity is actually higher at the edges and some googling told me it should more than compensate for that.

Constant Angular Velocity (http://en.wikipedia.org/wiki/Constant_Angular_Velocity)


The outer rings have less density because they're apparently more prone to errors, but faster speed, because of CAV.  The actual fastest sector is inwards of the outer edge of the disk.

I wonder if that explains the way that Windows 7 places the pagefile...just a bit in from the outer rim, and places the restore files as close to the axel as possible, since they aren't used routinely.

Exactly.  I am, for some incredibly dumb reason, interested in bringing some Real Knowledge forward about Windows... insofar as I have been pretty ignorant of such things for a long time.  These things we are talking about, I can do these and more in AIX and I really want to bring forth the same sort of ability against my home PC.  I might even go so far as to read things on MSDN or wherever such knowledge lives.

So, about the sorting of files to a particular sector... I need to read more of the thin documentation for MyDefrag but currently I know that it is possible to define zones and select certain files to fill that zone.  My current assumption is that Zone 1 is on the outer edge, so if I define a file selection that encompasses the OS files, those should be moved into Zone 1 and therefore to the outer edge.  Unfortunately I have been doing something <SCENE MISSING> and have not had time to practice my black arts on my own PC.

I'll make a tangent here and mention that a reasonable person/program would load oft-needed data into RAM and page out less-needed stuff.  Ideally I would like to read a texture only once, however in a game like The Sims 3 this is a massive challenge.

I would appreciate it if you would post interesting info that you find here.

I've never wanted to learn much about the specific handling of Windows simply because of the obscure and unmanageable nature of how their circular help docs, unintelligible tech information and just generally dumb ideas about UI make it an all around frustrating experience.

But, if you want to dig in up to the elbow and abstract out the nutty goodness I would be willing to learn and follow.

I spend a percentage of my time at work translating mind-numbing technical details into things that people who are not fluent in a topic can use to perform tasks... so yeah, I'll try to condense what I find.  The problem that recently cropped up is that my client/wife has grown impatient and wants me to pay a lot of money for whatever will fix the problem. :oh_i_see:  I will still practice on my own rig even if hers isn't available.

My experience with things that you buy to improve performance is that they generally don't and often can be counter-productive.

Which is why this thread was born. :grin: