Subject: Re: unified buffers and responsibility
To: Milos Urbanek <urbanek@openbsd.cz>
From: Wojciech Puchar <wojtek@chylonia.3miasto.net>
List: tech-kern
Date: 06/13/2002 12:04:38
> The response time of other apps is slow aswell during the
> copy/tar/whatsever 'larger' disk IO - including Window Manager, X itself, xterms
> and other apps, not only those like netscape.

yes everything is slowed down.

> during the command
> gzip -d netbsd.0.core in /var/crash
>
> imediately after the command completes the statistics are as follows:
> Memory: 62M Act, 23M Inact, 3904K Wired, 5084K Exec, 35M File, 18M Free
> Swap: 513M Total, 23M Used, 491M Free

18MB free because gz'ed file were deleted and it's buffer caches freed.

> vmstat -ms after the first command gzip netbsd.0.core.gz:
>
> In use 2889K, total allocated 3784K; utilization 76.3%
>
> I think the sysctl values for buff cache should be tuned a bit or is there another
> solution? Btw. the lock could be related to frequent swap ins/outs.

tuning them doesn't change reactions. even with vm.filemax=10 instead of
50!!!