Originally posted by: aidanjm
I have my main computer on 24-7, but I find if I don't reboot at least once a day, it becomes progressively more sluggish. Why?
I've noticed that too. I wonder how much of that has to do with system-managed memory heaps, combined with the "tweaks" that MS did to the Process Manager's working-set trimming algorithm. I know that MS has released some patches at various times to modify XP's heap algorithm, one of them was titled "patch for performance regression for atypical heap-allocation requests" or something like that. Basically, it seems like they fiddled with the code, and possibly broke a few things, or at least changed the observed performance behavior in terms of various cases, and apparently made some of the worst-case ones worse. I greatly prefer W2K's behavior in that regard, as while the best-case may have been a few theoretical performance percentage-points behind XP's current implementation, at least the performance was mostly deterministic, and didn't have the subjective performance behavior of "being sluggish at times". IMHO, at least.
Edit: Btw, I've noticed this myself, consistently, and it has nothing to do with spyware/malware, and it also doesn't have anything (that I can distinctly tell) has to do with a memory leak, because the VM size maxes out early on, and doesn't continue increasing. Yet the apps start to get slower and slower, by a slightly-perceptable margin. Rebooting of course brings everything up to full speed. I'm on a mid/low-range AThlon XP system, proper cooling, no thermal-throttling or power-manage operations going on.
I almost wonder if XP itself is "leaking" prototype PTEs or something. (Ok, totally wild guess, but not entirely inconsistent with my personal observations. So maybe a memory-leak is going on, but it's not an app leak of commited VM pages, but rather an OS-level one. I'm not entirely sure how system-managed heaps are handled, could severe fragmentation of those interact with PTE management in such a way as to cause these issues?)
I also wonder if video drivers could have a hand in this too, for example with my ATI AGP card, and running newest Prime95, for the last few released of the video drivers, I can't run the "Blend" test anymore, CPU time hovers at nearly zero, memory allocation remains low, and yet the HD goes nuts, thrashing like a mofo, because the OS is doing system-level PTE stuff allocating them for a certain VM size, and then paging out those PTEs to make room for more. I have also observed this behavior when using QuickPAR, and choosing too big a "chunk size" when creating par-sets. The app naively attempts to allocate that much VM offhand or something. One time I tried it with 10GB, unintentionally. Oops.
Edit: It happened again, opening Task Manager shows "Mem Usage" for all shown apps to total to roughly 105MB, this box currently has 256MB of RAM, and the HD is mostly constantly thrashing, even when I stop my file-copying/extracting/downloading for a bit. "Available memory" is ~5MB, "System Cache" is ~36MB. "Commit charge" is ~600MB or so. What I don't get is, if the current allocation of physical RAM to the working-sets of running processes is only 105MB, and the avail RAM/system cache is as stated, then where is the rest of the physical RAM going to? 150MB used up for non-pagable/kernel memory usage??? Considering that parts of the system are also pagable, this doesn't sound right at all. Unless there is some evilness going on with the Remote Desktop service's kernel-mode driver, and allocating bitmaps for all of the apps, and that's what is taking the room. But I'm sticking to my fragmented-VM/PTE mapping issue hypothesis, unless I find more evidence otherwise.