* Rik van Riel <riel@conectiva.com.br> [021130 02:57]:
>> root # vmstat 1
>> procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----
>> r b swpd free buff cache si so bi bo in cs us sy id wa
>> 0 1 265048 5248 32248 119108 6 15 65 62 252 585 21 6 73 0
>> 1 6 266648 4480 32316 120300 0 4656 2152 4652 1348 821 13 8 79 0
>> 1 0 265052 4496 31036 120184 8 336 1668 340 1226 765 15 7 78 0
>> 0 1 265052 4496 31112 121564 4 0 3152 0 1198 894 18 8 74 0
>> 1 0 265052 4504 31076 123112 0 0 3024 8576 1229 857 17 7 76 0
> ^^^^^^^^^^^^^^^^^^^
>Looks like my guess was right after all. The amount of swap
>IO is maybe 10% of the amount of filesystem IO in your vmstat
>snippet above.
>Two things could be happening here:
>1) the kernel decides to cache the wrong things in the
> page cache
>and/or
>2) the IO scheduler is giving worse latencies now
>If the problem is (1) it might get resolved by using the -rmap
>or -aa kernels. If the problem is (2) you'll want Andrew Morton's
>read_latency patch (which I'll port to 2.4.20 real soon now).
All right. I might be wrong, but this was with kernel 2.4.20-rc4-ac1
Doesn't it include rmap?
Also it was patched with Robert Love's preempt-kernel. Of course I've
tried without it as with various other kernel tidbits.
I always tend to use ac kernels + some other patches which aside from
preempt have nothing to do with system's basic behavior.
Yet the only time I get a good system response is with 2.5.47-ac6
(2.5.48+ drive me nuts with module loading) and with Marc-Christian
Petersen's 2.4.18-wolk which includes nearly everything out there.
-- Javier Marcet <jmarcet@pobox.com>
This archive was generated by hypermail 2b29 : Sat Nov 30 2002 - 22:00:23 EST