Re: dead mem pages

From: Dirk Wetter (dirkw@rentec.com)
Date: Wed Jul 11 2001 - 01:04:07 EST


Rik,

thx for your answer. :)

Rik van Riel wrote:

>On Tue, 10 Jul 2001, Dirk Wetter wrote:
>
>>>It would be good to know what these 2.8GB of cached pages are.
>>>
>>believe me, i would like to know too where all the $$$ memory
>>went to. ;-)
>>
>
>Most likely swap cache, that means it is the memory from your
>simulations, just removed from the page tables and put in the
>swap cache.
>
but why was the machine actually swapping then? sar definetely showed swap
and disk activity as the applications started.

>>>Again on a general note, the 2.4 kernel's VM is new and hence not fully
>>>mature. So the short and unhelpful answer to your query is probably that
>>>the current VM system is not well tuned for your workload (4.3GB of memory
>>>hungry simulations on a 4GB machine).
>>>
>>concerning the maturity that's also the answer i got from the kernel
>>guru's at last USENIX in boston. but ihmo it *should* become soon
>>better for the future if Linux intends to become bigger in the server
>>business. (my $0.02)
>>
>
>It'll get better as soon as we have the time, for 2.4.7
>the VM statistics have already improved a bit so people
>are no longer fooled by large "cached" figures ;)
>
Rik (and Wayne): it's *not only* the statistics. they were swapping like
crazy.
the only thing the machines were responding immediately to were icmp
packets.
no tcp/udp. keystrokes on the console were echoed 2 minutes after i
typed the command
in. with some patience i managed to execute "top" i caught pictures were
kswapd
was in the first line having 99% or so of one CPU and the load was between
20 and 30.

>
>Actual improvements to the code, if needed at all, will
>come with time ... more than $0.02 will get you ;)
>
not that i don't appreciate very much your work, but i had to learn that
improvements are
needed: we could swap our 4GB machines to death just by submitting jobs
in the size
of the ~physical memory to them. but i don't have any doubts that you
guys will manage
to do neccessary changes:-)

i do the profile tests Marcelo suggested (thx) and come back with some
numbers.

tschuess :-)

        ~dirkw

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Jul 15 2001 - 21:00:13 EST