Re: excessive swap-in time
- From: phil-news-nospam@xxxxxxxx
- Date: 17 Mar 2008 13:25:38 GMT
On Sun, 16 Mar 2008 18:05:28 -0700 (PDT) David Schwartz <davids@xxxxxxxxxxxxx> wrote:
| Another problem may be that if a process is idle for a long time and
| then becomes active, many of its pages may be the least recently used
| pages in the system. So when it starts causing a swap in, it may well
| swap itself out to get the pages in, resulting in an amusing musical
| chairs fiasco.
That is what I think may be going on. So how to avoid that? I think
the kernel needs to understand what (few) processes are most active,
at least up to a level that can fit in memory based on recent stats
for the process that show its active set count (as opposed to the total
count for all of its VM).
| You can think of obvious ways to tweak this to make this one case
| perform better, but you can think of cases that are made worse by
| those same tweaks.
Something has to be forced back out to swap in a process that becomes
active again. The question is which pages to select. Those of processes
that are recently inactive would be what ultimately would be swapped out.
Might as well go straight to them directly, rather than the newly active
process fighting against its own pages.
| Don't expect much help -- most programmers are working on efficient
| ways to use more memory rather than better ways to handle low memory
| situations. (Many typical machines now have 4GB-8GB and working sets
| that seldom exceed 1.5GB.)
And I have built a new machine with 8GB. As soon as I can get the video
card issues resolved on it, I can use it directly and see how it behaves.
FYI, it has ZERO swap space configured. I have done some loading on it
by running 4 parallel kernel compiles and it did quite well compared to
the machine I'm still using as my desktop.
As for the video issues, that should be a different thread and likely in a
different group. But the summary is I need text with instant switching.
I'm not using text all the time, but when I do, I need it that way. I have
not yet been able to get X to achieve that. I think the speed of machines
these days can and its more a matter of configuring X with full screen text
"windows" that are buffered in RAM or VRAM properly. That and some changed
semantics on mouse handling (so a mouse in an X based text window behaves
like a mouse in text console mode under gpm). The alternative to this is
to find a video card that is supported by both X (with high resultion in
24 bit color) and also SVGATextMode. I have ONE such video card, but it
is AGP based and new machines have PCIe instead of AGP. I don't know enough
about X to know where to reprogram it for this fast behaviour. There are
some slick window managers that do some fairly fast visual effects to move
between virtual desktops. But so far, the people I've talked to that use
them have been unable to figure out how to kill the effect and make it just
instantly switch, AND do it directly by mapping a single (Ctrl+Alt shifted)
keystroke to it. Someone who knows how to reprogram all the elements of X
would be a help here, but I can't find any that want to be helpful. So for
now I am stuck with getting a more modern video card that SVGATextMode can
still work on (or with hardware specs suitable to make SVGATextMode work).
--
|---------------------------------------/----------------------------------|
| Phil Howard KA9WGN (ka9wgn.ham.org) / Do not send to the address below |
| first name lower case at ipal.net / spamtrap-2008-03-17-0808@xxxxxxxx |
|------------------------------------/-------------------------------------|
.
- Follow-Ups:
- Re: excessive swap-in time
- From: Michel Talon
- Re: excessive swap-in time
- References:
- excessive swap-in time
- From: phil-news-nospam
- Re: excessive swap-in time
- From: Jan Knutar
- Re: excessive swap-in time
- From: phil-news-nospam
- Re: excessive swap-in time
- From: David Schwartz
- excessive swap-in time
- Prev by Date: Re: excessive swap-in time
- Next by Date: Re: Dynamic linking with LD_PRELOAD - get it compiled
- Previous by thread: Re: excessive swap-in time
- Next by thread: Re: excessive swap-in time
- Index(es):
Relevant Pages
|