Subject: Re: Limiting disk I/O?
To: Michael van Elst <email@example.com>
From: Jukka Marin <firstname.lastname@example.org>
Date: 11/13/2007 08:52:55
On Tue, Nov 13, 2007 at 07:40:09AM +0100, Michael van Elst wrote:
> What happened to http://netbsd-soc.sourceforge.net/projects/congest/ ?
Wouldn't it be simpler to implement a new sysctl'able variable to limit the
tps to some percentage of the disk I/O capacity? No need to monitor what
the processes are doing. I don't mind the processes hitting the buffer
cache as hard as they like as long as the real disk I/O (which is the
real bottleneck) is kept reasonable. (I don't know what happens if the
system runs out of buffers - I guess it may affect reading as well..)
Where does systat get the %busy value? Could the same value be compared
against the "vfs.sync.maxbusy" limit? If the user sets vfs.sync.maxbusy
to 90, then the syncer will adjust the tps count it generates such that
the disk(s) are only 90% busy at maximum. If vfs.sync.maxbusy==100, then
no limiting would be done.