Date: Mon, 26 Jun 2006 19:55:27 -0700 From: "Kip Macy" <kip.macy@gmail.com> To: "Robert Watson" <rwatson@freebsd.org> Cc: Perforce Change Reviews <perforce@freebsd.org>, Kip Macy <kmacy@freebsd.org>, John Baldwin <jhb@freebsd.org> Subject: Re: PERFORCE change 100089 for review Message-ID: <b1fa29170606261955r252e15a5l4ffc13d061dbef02@mail.gmail.com> In-Reply-To: <20060627001336.T79454@fledge.watson.org> References: <200606262054.k5QKsDq7022302@repoman.freebsd.org> <200606261759.41541.jhb@freebsd.org> <20060627001336.T79454@fledge.watson.org>
next in thread | previous in thread | raw e-mail | index | archive | help
I've mapped your uipc_usrreq.c into my tree and have seen a measurable
boost. I actually see no contention on it. If I go into overload (16
threads) I see the following:
65 13580255 555960120 4332486 3 128 22050892
4323043 /flatstor/shared/p4/sun4v/work_sleepq/src/sys/kern/kern_synch.c:217
(lockbuilder mtxpool)
13 24053476 160697931 92708398 0 1 30726211
0 /flatstor/shared/p4/sun4v/work_sleepq/src/sys/kern/kern_switch.c:522
(runq lock)
371 63389470 27487168 936871 67 29 5918460
640938 /flatstor/shared/p4/sun4v/work_sleepq/src/sys/kern/kern_lock.c:163
(lockbuilder mtxpool)
39 36405448 10970117 4748316 7 2 4132590
0 /flatstor/shared/p4/sun4v/work_sleepq/src/sys/kern/kern_switch.c:221
(runq lock)
361 85861725 10866103 5699832 15 1 3813907
0 /flatstor/shared/p4/sun4v/work_sleepq/src/sys/kern/subr_sleepqueue.c:223
(sleepq chain)
lockmgr is my biggest problem now.
On 6/26/06, Robert Watson <rwatson@freebsd.org> wrote:
> On Mon, 26 Jun 2006, John Baldwin wrote:
>
> > On Monday 26 June 2006 16:54, Kip Macy wrote:
> >> http://perforce.freebsd.org/chv.cgi?CH=100089
> >>
> >> Change 100089 by kmacy@kmacy_storage:sun4v_work_sleepq on 2006/06/26
> > 20:53:51
> >>
> >> add profiling for rwlocks
> >> not convinced of correctness as there don't appear to be any contended
> > rwlocks on my workloads
> >
> > Few things use them currently. I have a patch to make the name cache use
> > them if you want it.
>
> You may already have seen this, but I have a UNIX domain socket re-locking in
> //depot/user/rwatson/proto/src/sys/kern/uipc_usrreq.c that uses rwlocks and
> finer-grained mutexes, among other things. Ideally this can generate some
> contention (although perhaps not too much).
>
> Robert N M Watson
> Computer Laboratory
> University of Cambridge
>
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?b1fa29170606261955r252e15a5l4ffc13d061dbef02>