* 2.2.x series and mm
@ 2001-06-27 15:08 Adam
2001-06-27 15:20 ` Andrea Arcangeli
2001-06-27 16:27 ` Alan Cox
0 siblings, 2 replies; 7+ messages in thread
From: Adam @ 2001-06-27 15:08 UTC (permalink / raw)
To: linux-kernel
hello,
I have question. I have box with kernel 2.2.17pre15
and 128mb memory.
now on this box I have apache server which is serving 205 mb
of data.
AFAICT this casues all current processes swapped out every
so often in favor of putting all data to be served into
file buffers, making a box pain for interative use.
what are my options here for a) tuning up box w/out rebooting
or b) some "better" 2.2.x kernel.
I'm fairly sure it is the file buffers as the apache is already
reniced to 20, it is got max 50 processes and each of processes is
limited to like 1.5mb of size via ulimit.
--
Adam
http://www.eax.com The Supreme Headquarters of the 32 bit registers
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: 2.2.x series and mm
2001-06-27 15:08 2.2.x series and mm Adam
@ 2001-06-27 15:20 ` Andrea Arcangeli
2001-06-27 16:27 ` Alan Cox
1 sibling, 0 replies; 7+ messages in thread
From: Andrea Arcangeli @ 2001-06-27 15:20 UTC (permalink / raw)
To: Adam; +Cc: linux-kernel
On Wed, Jun 27, 2001 at 10:08:15AM -0500, Adam wrote:
>
> hello,
> I have question. I have box with kernel 2.2.17pre15
upgrade to 2.2.19 or 2.2.20pre
Andrea
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: 2.2.x series and mm
2001-06-27 15:08 2.2.x series and mm Adam
2001-06-27 15:20 ` Andrea Arcangeli
@ 2001-06-27 16:27 ` Alan Cox
2001-06-27 16:39 ` Adam
2001-06-28 7:33 ` Sean Hunter
1 sibling, 2 replies; 7+ messages in thread
From: Alan Cox @ 2001-06-27 16:27 UTC (permalink / raw)
To: Adam; +Cc: linux-kernel
> I'm fairly sure it is the file buffers as the apache is already
> reniced to 20, it is got max 50 processes and each of processes is
> limited to like 1.5mb of size via ulimit.
nice wont help you, it controls scheduling priority. Similar a ulimit just
ensures that no apache process goes mad and eats lots of memory (good idea
but not helpful here). If your working set (and thats the bit the matters)
really is exceeding memory by a fair bit then
a) Add more RAM - that is the real optimal approach
b) Make the processes smaller (eg switch to thttpd from www.acme.com)
c) Speed up the I/O throughput relative to CPU speed
- eg the 2.2 IDE UDMA patches
2.2.19+ do make slightly better decisions on the VM front, but at the end of
the day swapping only works usefully when the working set still fits in
RAM (ie all the stuff you keep needing).
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: 2.2.x series and mm
2001-06-27 16:27 ` Alan Cox
@ 2001-06-27 16:39 ` Adam
2001-06-27 16:52 ` Alan Cox
2001-06-28 7:33 ` Sean Hunter
1 sibling, 1 reply; 7+ messages in thread
From: Adam @ 2001-06-27 16:39 UTC (permalink / raw)
To: Alan Cox; +Cc: linux-kernel
> 2.2.19+ do make slightly better decisions on the VM front, but at the end of
> the day swapping only works usefully when the working set still fits in
> RAM (ie all the stuff you keep needing).
> a) Add more RAM - that is the real optimal approach
> b) Make the processes smaller (eg switch to thttpd from www.acme.com)
> c) Speed up the I/O throughput relative to CPU speed
> - eg the 2.2 IDE UDMA patches
can you elaborate on the "c" point" perhaps I could try it together with
2.2.20pre6 until I can do a).
about b) would it really help? AFACT the issue here is the buffers in
memory gets filled and cause other stuff to get swapped out., and that
would happen no matter what kind of web server I use..
--
Adam
http://www.eax.com The Supreme Headquarters of the 32 bit registers
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: 2.2.x series and mm
2001-06-27 16:39 ` Adam
@ 2001-06-27 16:52 ` Alan Cox
0 siblings, 0 replies; 7+ messages in thread
From: Alan Cox @ 2001-06-27 16:52 UTC (permalink / raw)
To: Adam; +Cc: Alan Cox, linux-kernel
> > b) Make the processes smaller (eg switch to thttpd from www.acme.com)
> > c) Speed up the I/O throughput relative to CPU speed
> > - eg the 2.2 IDE UDMA patches
>
> can you elaborate on the "c" point" perhaps I could try it together with
> 2.2.20pre6 until I can do a).
>
> about b) would it really help? AFACT the issue here is the buffers in
> memory gets filled and cause other stuff to get swapped out., and that
> would happen no matter what kind of web server I use..
It depends if it takes the working set down (you dont care about the total
amount of data but the amount regularly being used) - thttpd uses a _lot_
less memory for the webserver itself so can help
As to c) - 2.4.x and with patches 2.2.x will do UDMA66/UDMA100 I/O on modern
disks and that takes the CPU usage down (so you do more work while the
disk is copying stuff) and might well be getting data on/off disk ten times
as fast
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: 2.2.x series and mm
2001-06-27 16:27 ` Alan Cox
2001-06-27 16:39 ` Adam
@ 2001-06-28 7:33 ` Sean Hunter
2001-06-28 9:52 ` Oliver Teuber
1 sibling, 1 reply; 7+ messages in thread
From: Sean Hunter @ 2001-06-28 7:33 UTC (permalink / raw)
To: Adam; +Cc: linux-kernel
On Wed, Jun 27, 2001 at 05:27:11PM +0100, Alan Cox wrote:
> > I'm fairly sure it is the file buffers as the apache is already
> > reniced to 20, it is got max 50 processes and each of processes is
> > limited to like 1.5mb of size via ulimit.
>
> nice wont help you, it controls scheduling priority. Similar a ulimit just
> ensures that no apache process goes mad and eats lots of memory (good idea
> but not helpful here). If your working set (and thats the bit the matters)
> really is exceeding memory by a fair bit then
>
> a) Add more RAM - that is the real optimal approach
> b) Make the processes smaller (eg switch to thttpd from www.acme.com)
> c) Speed up the I/O throughput relative to CPU speed
> - eg the 2.2 IDE UDMA patches
It may also be worth considering
d) Reduce the number of Apache processes so they fit nicely in RAM
Sean
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2001-06-28 9:53 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2001-06-27 15:08 2.2.x series and mm Adam
2001-06-27 15:20 ` Andrea Arcangeli
2001-06-27 16:27 ` Alan Cox
2001-06-27 16:39 ` Adam
2001-06-27 16:52 ` Alan Cox
2001-06-28 7:33 ` Sean Hunter
2001-06-28 9:52 ` Oliver Teuber
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox