RE: Exaggerated swap usage

[Randal, Phil]

Same thing happens with RedHat 8.0's latest 2.4.18-18 kernel.
Box has been up for almost 8 days.

top:

Mem:  1031036K av, 1021140K used,    9896K free,       0K shrd,   80560K
buff
Swap: 1052216K av,    5064K used, 1047152K free                  627808K
cached

vmstat:

   procs                      memory    swap          io     system
cpu
 r  b  w   swpd   free   buff  cache  si  so    bi    bo   in    cs  us  sy
id
 0  0  0   5064  10192  80860 626932   0   0     2    21    3    56   0   0
36

Swapping in these circumstances shows a pathological reluctance to shed
cached pages.

Not critical, but hardly optimal behaviour.  Or is it?

Phil
---------------------------------------------
Phil Randal
Network Engineer
Herefordshire Council
Hereford, UK 

> -----Original Message-----
> From: Javier Marcet [mailto:[e-mail]]
> Sent: 29 November 2002 11:54
> To: [e-mail]
> Subject: Exaggerated swap usage
> 
> 
> 
> Forgive me if I don't provide enough information just yet, or am not
> clear enough. I simply don't know what setting to tweak.
> 
> I'll explain.
> In recent 2.4.20 pre and rc kernels ( I tend to use the ac branch ), I
> had notice my system, when using X mainly, got terribly slow 
> after some
> use. It surprised me that when I tried 2.5.47 this did not happen at
> all, since I thought my problem was a lack of memory - the system has
> 384MB -.
> Hence I tried to find where the difference was. What I found is that
> 2.4.20 kernels - 2.4.19 does the same -, was swapping just too much,
> while there was a lot free memory on the system, cached but free.
> I disabled all swap and it suddenly began to work smoothly again, yet
> with the random kills when memory was a scarce resource on the system.
> I've tried different sysctl's vm.overcommit settings but the result is
> the same.
> 
> I also found a 2.4.x kernel which did not show this behavior, WOLK, in
> any version I tried.
> 
> Could you please point me toward something I can try tweaking, or some
> documentation to read which explains what I can change, 
> unless it's some
> kind of kernel problem?
> 
> BTW, aa kernels behaved somewhat better on this, only that 
> the last one
> I tried -rc2aa1- had some stability problems.
> 
> I can provide you with dmesg, /proc/meminfo or whatever might 
> be useful.
> 
> Thanks in advance :)
> 
> 
> -- 
> Javier Marcet <[e-mail]>
>