Is this true about Vista?

[BD2003]

Originally posted by: andy04

Originally posted by: BD2003

Originally posted by: harpy82
anyone knows a good way or benchmark to actually test this?? I do think a fast flashdrive might help a little since seek time are 0.00sec compare to a harddisk swapfile.

It's very, very difficult to benchmark.

You can't just run a test, pop the stick in and run it again, since it's fairly useless until things have been cached on it.

The persistent memory cache also complicates things a great deal, since you'd have to flush that cache out, or else the ram disk cache will obviously blow away the flash disk.

I kept on eye on it with perfmon, and it doesnt do much when you're not running low on memory...there's little swapping, and you have a decent ram disk cache which is clearly your first choice.

Any benchmarks I've found on the net miss these points entirely, and write it off as useless. Maybe if I have some free time later today, I'll run a few of my own.

I did see major activity when I was loading games and such - things that really tax your memory so you're not only low on program memory, but have a tiny ram disk cache. This is why its so commonly understood as a replacement for memory - even though its just a disk cache, it really provides it's main benefit under low memory situations.

So if you have 2gb of memory or so, and rarely fill it, and are expecting a significant boost, you're not going to find it.

Another thing that will impact the benchmark i think will be the simultaneous use of an external USB HDD... Just a thought... (no need to get anal if i am wrong... ppl in this thread have been acting pretty weird...)

Why would that impact it? Unless youre talking about using a usb HDD for readyboost...which you couldnt do; it'll never pass the test because it would be utterly useless.

 

[Nothinman]

Another thing that will impact the benchmark i think will be the simultaneous use of an external USB HDD... Just a thought... (no need to get anal if i am wrong... ppl in this thread have been acting pretty weird...)

If you're thinking about the flash drive and the regulard USB hard disk sharing USB bandwidth and slowing each other down, yes I believe that's possible but the flash drive being used by ReadyBoost won't be having any long sustained transfers from it so I doubt it'll make a noticable difference. And on top of that the data cached on the flash is just that, a cache. The data's also either in memory or on another hard disk so Windows can go get the data from either device.

 

[BD2003]

An informal test, far from scientific, and far from completely controlled:

The test is how long is takes to load COD2 to the initial intro movie. Specs are p4 3.0ghz, 2gb ram, 7200rpm drive.

One important thing to note is that I have the swapfile on a separate drive from windows/games, in a dedicated partition on the fastest part of the drive. This already makes a fairly huge difference since SATA bandwidth isnt shared, the heads on the drives dont have to swing back and forth between the swap file and the game file etc...for those of you with a single drive, the HD has to do a lot more work to swap.

Readyboost not enabled yet.

This is the first install of the game, so theres no chance superfetch could have it. On a cold boot: 24 seconds

Immediately reloading after that, while still in ram cache: 10 seconds.

Next, I closed the game, and filled up the ram with huge 400mb bitmaps - from the perfmon, ram was filled, cache was completely nonexistent and it began swapping. No matter what I did, after loading the bmp, vista made sure to keep 80mb of disk cache...what was in there...no way to tell, but I doubt anything from COD survived.

Once swapping ceased, I reloaded: 36 seconds, with the disk access to the swap partition going wild.
Closed the game out, and immediately reloaded: 16 seconds - not as good as the 10 seconds before, since ram was still limited.

Then I enabled readyboost, and cleared the cache again by loading the bmps.

Loading the game: 36 seconds - lots of swapping, and noticible writing to the readyboost drive as well. As expected - no data is cached for cod in the RB drive yet.
Immediate reload afterwards: 14 seconds. Mostly due to the ram cache, but the RB cache may have helped out somewhat.

Now that the data is definitely cached in the USB stick, I reloaded bmps again to clear the ram cache.

Load: 22 Seconds. Quite a difference I'd say. Still a fair bit of swapping, but since some of the data could come from the stick, it was able to make up for some of it.
Immediate reload afterwards: 14 seconds. If it's in the ram cache, its in the ram cache, so RB isn't going to lower this number much.

All that being said, the test is still a hackjob, so don't put too much stock in it.

A better test would be to have 512/1gb ram isntead of using bmps to flush memory, and keeping the swapfile on the same drive as the game/OS, like most systems are configured. Then you'd likely see even bigger differences.

Also, the USB stick I'm using *barely* passes the performance tests...using a faster stick would probably give even better results.

It's VERY hard to have a controlled benchmark in Vista. There are too many factors that you can't control, mostly due to superfetch. It's interesting to watch that work though. When you boot, it loads the OS files and startup programs. After thats done, a few seconds later, the HD kicks in again, but it's preemptively loading things into the disk cache (which still appears as "free" memory) - since free memory isnt going down, but disk access is still occuring, it seems like vista is just accessing the disk for the hell of it, and gives the impression of disk swapping when it's doing the exact opposite.

Not only that, but you can see where it loads the files from - on the next boot after I did those tests, I noticed it was reading from my games partition. So after it was done (took a minute or two to fill ram), I fired up COD2, and it only took 18 seconds to load, so it was ready and waiting for me in the ram cache, right off a boot. Thats quality stuff right there. Yet that data is still in "free" memory, so its ready to be dumped if programs actually require all that memory.

 

[Nothinman]

Thats quality stuff right there. Yet that data is still in "free" memory, so its ready to be dumped if programs actually require all that memory.

That's how the filesystem cache has always worked, the only change in Vista is the preemptive paging in of data that you used the last time Vista booted.

 

[BD2003]

Originally posted by: Nothinman

Thats quality stuff right there. Yet that data is still in "free" memory, so its ready to be dumped if programs actually require all that memory.

That's how the filesystem cache has always worked, the only change in Vista is the preemptive paging in of data that you used the last time Vista booted.

Indeed, I mention that only to clarify things to people that might not be aware of that - "free memory" reported by windows is actually useful cache data the amount of truely free memory is usually only a few mb- the system cache line in the task manager is *not* the disk cache.

But the premptive paging is a pretty big deal IMO. Since I installed vista, just about every program I launch comes directly out of memory, the first time, and makes for a much snappier system.

And it doesnt just work on boot, it works continuously. Once I cleared out all of those bmps, it went back to work preloading everything into the cache.