Subject: Re: SMP re-entrancy in kernel drivers/"bottom half?"
To: None <tech-kern@netbsd.org>
From: Christos Zoulas <christos@tac.gw.com>
List: tech-kern
Date: 02/23/2005 15:50:42
In article <E1D42w2-0002cV-00@smeg.dsg.stanford.edu>,
Jonathan Stone  <jonathan@dsg.stanford.edu> wrote:
>In message <20050223195822.GN388@canolog.ninthwonder.com>,
>Allen Briggs writes:
>
>>I think there has been some discussion of this recently on a couple
>>of different NetBSD mailing lists.  Perhaps netbsd-advocacy@ and
>>tech-perform@ ?
>
>Hm. Indeed, the tech-perform@ archives do have discussion.  Turns out
>these same benchmark results triggered Chuck's finishing up read()
>loanout for regular files (so he can do write-loanout, which will help
>overwrite (a` la bonnie rewrite?), which is expected to help the
>10M-row tests where NetBSD does so poorly.
>
>I also notice PTHREAD_CONCURRENCY being changed from the old default
>of 1 to the number of CPUs found at boot, which was expected to
>dramatically improve the two-CPU tests. (Though I dont see why that
>would make us strictly worse with two CPUs versus one CPU?)

I reverted PTHREAD_CONCURRENCY back to 1, until the issues with it
have been fixed.

christos