View Full Version : Variations in performance...
vagreys
04-26-2007, 12:24 PM
I have noticed that, sometimes, if I shutdown and restart FAH, I can get a 75% or better improvement in performance. For example, working on the supervillin-3 series of work units (5000 frames), if I see that it is processing at 1m52s/frame, I can shutdown and restart FAH, and the processing time will drop to something in the neighborhood of 36s/frame or faster, changing nothing else on my machine including open applications. Any ideas why this would be?
Anything running in the background? Something that's activated during 'idle' ?
Taeric
04-26-2007, 01:13 PM
What are the specs on your system?
How long does the performance gain last? Is it just for a percent or two, or does it last for a while?
I have some suspicions, but I'll reserve my guess until I have a bit more info.
vagreys
04-26-2007, 02:18 PM
Anything running in the background? Something that's activated during 'idle' ?
Nothing that I'm aware of.
What are the specs on your system?
How long does the performance gain last? Is it just for a percent or two, or does it last for a while?...
1.6GHz Pentium M, 512mb RAM
For FAH, Core Priority is lowest possible, CPU Usage is max, checkpoint at 15 min, no power saving.
The change in performance lasts until the next shutdown, as far as I can tell. I realize it fluctuates some with other demands on the CPU, but most of the time, the machine is largely idle, since most of my work is being done on remote hosts.
Taeric
04-26-2007, 02:51 PM
My best guess is that the slowdowns are a result of Windows not doing a good job reclaiming RAM once an application is closed. If you're running XP or Vista, the operating system itself is a resource hog, and the folding client can use anywhere from just a few MB of RAM up to around 300 MB on certain WUs (100 MB is fairly common). If a you run a few applications here and there and the RAM used by those applications is not fully "returned" to the system, RAM could get to be a major choke point. Restarting cleans the slate, though, which may account for the change in production rate.
It would be interesting to run a RAM cleaner when the system is sluggish to see if that is indeed the issue. While it is indeed a good idea to restart a system from time to time, such a program may extend the time between restarts.
vagreys
04-26-2007, 03:09 PM
...It would be interesting to run a RAM cleaner when the system is sluggish to see if that is indeed the issue. While it is indeed a good idea to restart a system from time to time, such a program may extend the time between restarts.
I'm running under XP Pro SP2. I am not restarting the system, just the folding client. I've wondered if it had something to do with the resources available during startup, since XP runs a boatload of stuff at startup, while the folding client is also starting. After things have settled down, some, if I shutdown and restart the folding client, it starts processing frames much faster.
Taeric
04-26-2007, 03:30 PM
Hrm...my misunderstanding. I can't think of much of a reason in that case unless it has to do with how RAM is allocated between the physical memory and swap file. As you said, startup is a busy time, and the folding client may well be working party off of the RAM and partly off of the much slower virtual memory. XP should reallocate the physical memory to programs when there's enough available, but maybe it is doing so imperfectly.
If something as simple as restarting the folding client helps, though, we could speculate all day without coming up with a better solution. Do keep in mind that not ever frame within a given work unit is created equal. There will be some variation in times within many WUs, potentially significant variations, though they will not tend to last all that long (no more than a few percent).