Subject: Re: locking problem on amd64 (w/ lfs)
To: Antti Kantee <firstname.lastname@example.org>
From: Blair Sadewitz <email@example.com>
Date: 12/03/2007 16:27:33
I've found that LFS is more unstable if one uses non-default
lseg/block/frag sizes when creating the filesystem. At this time, I
recommend sticking to 1MB/8k/1k. Also:
-- Don't disable vfs.lfs.ignore_lazy_sync; if you do, you'll likely
end up with vnodes stuck in IN_PAGING. This is not fun and makes it
hard to sync disks upon shutdown.
-- Use vfs.lfs.pagetrip with caution (this might apply to MP systems
only). The lower the value, the more likely it seems that you'll end
up with a locking assertion or something else. There is some funky
problem when using this (seems to always involve the wakeup of the
writer daemon, IIRC, though I can't seem to find the backtrace I wrote
down last time this happened, so I'm not sure). It may be related to
the problem ad@ described.
> I use it for /obj all the time, no tweaking, works great.
Is this on a multiprocessor system? Me too, so long as I don't
deviate from the defaults it likes in any way. Also, I find that for
some reason that completely escapes me, it is more unstable as a root
filesystem. Why?!? ;)