Subject: Re: UVM/other problems for desktop users in current?
To: Peter Seebach <email@example.com>
From: Daniel Carosone <firstname.lastname@example.org>
Date: 12/18/2002 15:49:40
This is mostly a CVS issue, or rather the impact of CVS's access
For those experiencing this problem - are you running softdep?
CVS hits lots and lots of files, that means a lot of atime updates
in the first instance, and a bunch of other directory and inode
updates as files are actually pulled down. Softdep smoothes the
load of this considerably, and probably eliminates a lot of duplicate
updates. It can also help some of the other apps not block on their
own disk updates for so long while there's disk contention going
People have made the contrast between NetBSD and FreeBSD/Linux.
FreeBSD is probably using softdep by default (?), Linux is effectively
mounting async. Do the same on NetBSD, and then compare again.
Other things to help further:
- On my laptop, in order to let the disk stay spun down more of
the time, I run filesystems noatime,nodevmtime, further reducing
the load and impact of CVS. With the combination of all of
these things, I really see nothing like the problems people
describe here -- even though the disk traffic is all going
through cgd doing crypto in interrupt context.
- In current, there is work under NEW_BUFQ_STRATEGY to address
some of the other known problems with dealing with the impact of
nasty disk access patterns. It might be worth seeing what effect
this has, if you're willing to experiment. I haven't needed to.