Re: Performance benefits of rep/loop?
From: Matt Taylor (para_at_tampabay.rr.com)
Date: 05/31/04
- Next message: Robert Redelmeier: "Re: MOVNTQ on Athlon"
- Previous message: Matt Taylor: "Re: No need to optimize in assembly anymore"
- In reply to: Ivan Korotkov: "Re: Performance benefits of rep/loop?"
- Next in thread: Matt Taylor: "Re: Performance benefits of rep/loop?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Mon, 31 May 2004 01:51:24 +0000 (UTC)
"Ivan Korotkov" <koroNOSPAMtkov2@ztelDOT.ru> wrote in message
news:c98dk1$2j3i$1@news1.macomnet.ru...
> > Page stealing or whatever you'd like to call it is a good idea. Evicting
> > files from the cache to keep more than 1 GB of ram free is utterly
stupid.
> > They don't even try to predict the paging demand. I turn off the swap
file
> > as well to avoid this undesirable behavior.
>
> But Windows still is able to remove image pages from RAM and page-in them
> again, isn't it? And furthermore, doesn't it's behaviour mainly end up
with
> just moving pages between stand-by and ready lists without actually
> "page-inning" and "page-outting" to/from HDD?
Yes, or at least I presume so; otherwise I would expect that page faults did
not ever occur, but Windows Explorer continues to fault about once per
second. However, modified pages are now forced into memory since the OS has
nowhere to put them. That seems to dramatically improve performance of most
apps. The delay when switching between applications used to be quite
noticible.
The main problem is that Windows is too aggressive when trying to trim
working set sizes. It will look for victim pages even when there is no
logical need for it (e.g. 200-300 MB commit charge on a machine with 2+ GB
of ram). It won't use that memory for the file cache necessarily, either; it
keeps a number of pages ready to be allocated, and I suspect this number is
proportional to the amount of system memory.
-Matt
- Next message: Robert Redelmeier: "Re: MOVNTQ on Athlon"
- Previous message: Matt Taylor: "Re: No need to optimize in assembly anymore"
- In reply to: Ivan Korotkov: "Re: Performance benefits of rep/loop?"
- Next in thread: Matt Taylor: "Re: Performance benefits of rep/loop?"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
|
