Date: Tue, 8 Jul 2014 15:46:36 -0400 (EDT) From: Rick Macklem <rmacklem@uoguelph.ca> To: Glen Barber <gjb@FreeBSD.org> Cc: Rick Macklem <rmacklem@FreeBSD.org>, freebsd-stable@freebsd.org, =?utf-8?Q?Jos=C3=A9_Mar=C3=ADa?= Alcaide <jose@we.lc.ehu.es> Subject: Re: 9-STABLE nfsd(8) does not auto-tune number of threads (error in the 9.3 Release Notes!) Message-ID: <1539064411.8855953.1404848796119.JavaMail.root@uoguelph.ca> In-Reply-To: <20140708133559.GF1216@hub.FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
Glen Barber wrote: > On Tue, Jul 08, 2014 at 03:06:38PM +0200, José María Alcaide wrote: > > One of the 9.3-RELEASE Release Notes (userland changes) states > > that: > > > > "The default number of nfsd(8) threads has been increased from 4 to > > (8 * N), where N is the number of CPUs as reported by sysctl -n > > hw.ncpu. [r262124]" > > > > Indeed, revision r262124 > > (http://svnweb.freebsd.org/base?view=revision&revision=262124) > > removes the "-n 4" option from nfs_server_flags in > > /etc/defaults/rc.conf. > > > > The change is based on the premise that nfsd(8) auto-tunes the > > number of threads as (8 * hw.ncpu). That auto-tuning was > > introduced in HEAD revision r243637 of nfsd.c > > (http://svnweb.freebsd.org/base?view=revision&revision=243637), > > but it has not been MFC'd to 9-STABLE yet, so that release note > > *is not true*. > > > > I think that r262124 should be reverted, also removing the release > > note mentioned above; or alternatively bring HEAD's nfsd.c on sync > > with 9-STABLE (probably a bad idea just before 9.3-RELEASE). > > > > Reverting r262124 is also not an option at this point of the release > cycle. > > Rick, do you have any reservations to issuing an errata notice for > this > after the release? > Well, I think that it will generate 4 threads without the "-n 4" just like it does with the "-n 4", so the only issue I see is the statement in the release notes isn't accurate. (It doesn't seem worth reverting r262124, since it doesn't really affect the outcome?) I'd suggest just taking the statement out of the release notes. rick ps: This change wasn't my commit, but I didn't see a problem with the default changing. I have fixing the nfsd(8) man page to reflect this change in default on my "to do" list. > Glen > >
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?1539064411.8855953.1404848796119.JavaMail.root>