vista sp1 memory bug?

[blanketyblank]

Installed SP1 yesterday using live update, and my memory seemed really screwy.
I noticed incredible delays as if my CPU was stressed, but it tunrs out that what was being all used up was my RAM. I only use 2 GB with vista 32, which is usually enough. However when it uses 93% of my RAM as I saw it was while I'm just surfing the internet the computer comes to a crawl.
I tried disabling superfetch to see if that would free up some RAM, but even then I only have about 1.1 gigs free. I know Vista likes to take up a good chunk, but over 512 MBs is just ridiculous.

I thought I might have a virus that somehow got installed along with sp1 so I tried downloading zone alarm. BIG Mistake. The installation for that stopped halfway and seemed to cut off my ability to connect to the internet.

Finally I just gave up and restored an earlier state. Now I pray that windows update doesn't automatically install SP1 on me when I go to sleep...

 

[postmortemIA]

this is not a SP1 issue, superfetch did that all the time it was enabled from the day one...

window manager (aero theme) and OS will take ~800MB if you have them available
if you have enough of junk, there goes the rest

 

[Piuc2020]

Like someone else said on this forum, free memory is wasted memory so don't worry about it unless your Vista and Applications are running worse.

 

[blanketyblank]

I know that superfetch uses up my free memory to load stuff. I leave it enabled normally, but I disabled it because I figured SP1 was fetching too much stuff and not giving the memory back. I have a quadcore so normally I don't see any delays when I minimize/maximize a window. However because something was using up almost all of my memory it felt like I was on a single core currently running prime95.

It could be windows update, so I'll probably try downloading the standalone to see if that works for me. Could potentially be the update was installed while I was overclocked to 2.790 Ghz though that is 24hour prime stable for me.

 

[flexy]

i would be concerned if i had 4gb ram and only 10% of it is used all the time

 

[BD2003]

Originally posted by: blanketyblank
I know that superfetch uses up my free memory to load stuff. I leave it enabled normally, but I disabled it because I figured SP1 was fetching too much stuff and not giving the memory back. I have a quadcore so normally I don't see any delays when I minimize/maximize a window. However because something was using up almost all of my memory it felt like I was on a single core currently running prime95.

It could be windows update, so I'll probably try downloading the standalone to see if that works for me. Could potentially be the update was installed while I was overclocked to 2.790 Ghz though that is 24hour prime stable for me.

Superfetch will entirely fill your memory until its nearly 100% full. This is a *good* thing. It gives it all back instantaneously when another program needs it, its cache just like it was in XP, the only difference being that Vista is a lot more "honest" about the ram it uses. XP considered cache "free memory" in the task manager, Vista only considers empty memory as "free". Its just a semantic difference, in either case, its usually almost 100% full of stuff, the only difference being the stuff managed by Vista is more likely to be useful.

 

[BD2003]

Either way, after an install of updates or a service pack, the next boot usually takes usually long, but should go away on the boot or two after that.

 

[jdoggg12]

Originally posted by: flexy
i would be concerned if i had 4gb ram and only 10% of it is used all the time

IMO that situation is like buying a helmet and being pissed you dont use it enough - most functions of computing wont use anywhere near 4Gb of ram, but for the few things that do, you'll love it.

Keep in mind this is from a gamers standpoint

 

[masteraleph]

Originally posted by: jdoggg12

Originally posted by: flexy
i would be concerned if i had 4gb ram and only 10% of it is used all the time

IMO that situation is like buying a helmet and being pissed you dont use it enough - most functions of computing wont use anywhere near 4Gb of ram, but for the few things that do, you'll love it.

Keep in mind this is from a gamers standpoint

Except that as soon as you do something that requires the ram, windows lets it go. From a gamer's standpoint, it doesn't matter at all.

 

[jdoggg12]

Dude, you read what i said entirely wrong.

I said that gamers PREFER to have the system using as little ram as possible for when ram intensive apps need it, its there. My argument had nothing to do with superfetch.

 

[Snapster]

Originally posted by: jdoggg12
Dude, you read what i said entirely wrong.

I said that gamers PREFER to have the system using as little ram as possible for when ram intensive apps need it, its there. My argument had nothing to do with superfetch.

And they do get it, Windows literally dumps the areo theme and goes into windows 'basic' mode and frees up all the necessary resources for the game.

 

[kylef]

Disabling SuperFetch is a really bad idea. SuperFetch exists only to make the Cache Manager work more efficiently. Among many other functions, for example, it prevents apps which crawl the filesystem (like virus scanners, indexers, etc) from evicting cached pages for apps you actually care about. It does this by analyzing the usage patterns of processes and giving hints to the cache manager about prioritization.

And aside from protecting often-used pages from cache eviction, it also (like Prefetch) keeps lists of "often-used" pages and speculatively loads them from disk when the system has lots of available physical pages. It assists in fast recovery from shutdown and hibernate. And while you might think that such speculative IO would cause slowdowns by tying up your disk, from what I understand Superfetch makes extensive use of low-priority I/O (a new Vista feature) to prevent these types of optimizations from bothering the "important" disk I/O of user processes like IE, Word, Media Player, etc.

There are lots of things you can turn off in Vista which will help overall performance (the Search Indexer and Windows Defender are two prime examples), but SuperFetch should NEVER be placed in this category IMHO.

 

[BD2003]

Originally posted by: kylef
Disabling SuperFetch is a really bad idea. SuperFetch exists only to make the Cache Manager work more efficiently. Among many other functions, for example, it prevents apps which crawl the filesystem (like virus scanners, indexers, etc) from evicting cached pages for apps you actually care about. It does this by analyzing the usage patterns of processes and giving hints to the cache manager about prioritization.

And aside from protecting often-used pages from cache eviction, it also (like Prefetch) keeps lists of "often-used" pages and speculatively loads them from disk when the system has lots of available physical pages. It assists in fast recovery from shutdown and hibernate. And while you might think that such speculative IO would cause slowdowns by tying up your disk, from what I understand Superfetch makes extensive use of low-priority I/O (a new Vista feature) to prevent these types of optimizations from bothering the "important" disk I/O of user processes like IE, Word, Media Player, etc.

There are lots of things you can turn off in Vista which will help overall performance (the Search Indexer and Windows Defender are two prime examples), but SuperFetch should NEVER be placed in this category IMHO.

Pretty much. Its well designed - there are two minor issues with it that come nowhere near outweighing its benefits.

1. It uses I/O priority, which works relatively well. But since we're dealing with spinning discs, due rotational latency and seek times, low priority i/o can be only so effective. Once we're only solid state discs, this problem ceases to exist, low priority i/o will work perfectly. Until there, there will be a relatively slight performance hit *only while SF is filling the cache*. This really only happens on boot and after closing large programs like games. The disk thrashing makes it sound like its slowing the PC down more than it really is thanks to low priority I/O, but make no mistake, there is still a slight hit during those few seconds, but you would probably have to benchmark it to tell.

2. The service monitoring and containing the information on how to manage the cache takes up a decent amount of memory on its own, which is dependent on how much memory it needs to manage. On my 4gb system, it can soak up nearly 150mb, on my 1gb system, just 24mb. But even so, thats memory well used and put to a good purpose.

But there is no performance hit once its loaded. Itll drop cache as quick as XP did, which is to say, instantaneously, the same as if it was truely free memory. I can think of very, very, very, very few scenarios where you'd even want to consider turning it off.

 

[kylef]

Originally posted by: BD2003
2. The service monitoring and containing the information on how to manage the cache takes up a decent amount of memory on its own, which is dependent on how much memory it needs to manage. On my 4gb system, it can soak up nearly 150mb, on my 1gb system, just 24mb. But even so, thats memory well used and put to a good purpose.

Just curious, but how were you measuring the memory footprint of the SysMain service by itself? Did you force it to start in its own svchost process?

On my 64-bit 4GB system, the svchost.exe process which includes SuperFetch and 11 other services (e.g., Netman, the audio service, the client-side caching service, etc) is using a total of 136 MB of private memory, after an uptime of 13 days. i'm not convinced that Superfetch is responsible for all, or even most, of that overhead.

But there is no performance hit once its loaded. Itll drop cache as quick as XP did, which is to say, instantaneously, the same as if it was truely free memory. I can think of very, very, very, very few scenarios where you'd even want to consider turning it off.

Agreed. Regarding "dropping cache", it helps to remember that SuperFetch itself is not the party responsible for reassigning pages. That role still belongs solely to the Cache Manager itself, and the algorithm used in Vista is very similar to that used in XP from what I understand, the key difference being that Vista's Standby List now has multiple prioritized levels, and the lowest priority lists are evicted first. SuperFetch plays a role in ensuring pages are assigned the right priority level so that "important" pages are never swapped to disk due to cache pressure caused by "unimportant" process behavior.

Also, regarding Free vs. Cached pages. Moving pages from the Standby list to a requesting process's address space is slightly more expensive than moving from the Free list, due to the need to scrub all data from the page first (a basic process-level security feature). Pages sitting in the Free list have already been scrubbed by the background "Zero page thread" and are literally ready to go immediately.

But keeping Free list pages scrubbed of data also means that they are useless to the system until they are re-assigned. So the OS tries to strike a balance, keeping the vast majority of "freeable" pages on the system in the Standby list where they can serve a purpose, and a much smaller portion zeroed and available instantly. I believe the Zero Page thread wakes up periodically and enforces this balance, evicting enough low-priority standby list pages and scrubbing them to keep the Free list at its target size.

 

[BD2003]

Originally posted by: kylef

Originally posted by: BD2003
2. The service monitoring and containing the information on how to manage the cache takes up a decent amount of memory on its own, which is dependent on how much memory it needs to manage. On my 4gb system, it can soak up nearly 150mb, on my 1gb system, just 24mb. But even so, thats memory well used and put to a good purpose.

Just curious, but how were you measuring the memory footprint of the SysMain service by itself? Did you force it to start in its own svchost process?

On my 64-bit 4GB system, the svchost.exe process which includes SuperFetch and 11 other services (e.g., Netman, the audio service, the client-side caching service, etc) is using a total of 136 MB of private memory, after an uptime of 13 days. i'm not convinced that Superfetch is responsible for all, or even most, of that overhead.

Simple. Using task manager, I located the svchost.exe that the superfetch runs on. I then stopped the superfetch service, and looked for its effect. Just did it again right now on my 4gb system

SF On - The svchost.exe in question is using 155mb.
SF Off - Drops to 34mb instantaneously.
SF On again - Back to 133mb instantaneously.

Agreed. Regarding "dropping cache", it helps to remember that SuperFetch itself is not the party responsible for reassigning pages. That role still belongs solely to the Cache Manager itself, and the algorithm used in Vista is very similar to that used in XP from what I understand, the key difference being that Vista's Standby List now has multiple prioritized levels, and the lowest priority lists are evicted first. SuperFetch plays a role in ensuring pages are assigned the right priority level so that "important" pages are never swapped to disk due to cache pressure caused by "unimportant" process behavior.

Also, regarding Free vs. Cached pages. Moving pages from the Standby list to a requesting process's address space is slightly more expensive than moving from the Free list, due to the need to scrub all data from the page first (a basic process-level security feature). Pages sitting in the Free list have already been scrubbed by the background "Zero page thread" and are literally ready to go immediately.

But keeping Free list pages scrubbed of data also means that they are useless to the system until they are re-assigned. So the OS tries to strike a balance, keeping the vast majority of "freeable" pages on the system in the Standby list where they can serve a purpose, and a much smaller portion zeroed and available instantly. I believe the Zero Page thread wakes up periodically and enforces this balance, evicting enough low-priority standby list pages and scrubbing them to keep the Free list at its target size.

Indeed. But with the speed of memory nowadays, its probably far, far beyond the bounds of human perception to notice any lag caused by clearing out a chunk of memory.

As of right now on my system, theres 1540mb in use, 254mb in non-standby cache (recent file I/O cache), 2255mb in standby (superfetch managed cache) and 41mb free. It generally maintains this balance of the "free" memory - a small chunk for the recent cache, a tiny chunk kept zeroed out, and the vast majority managed by SF.