Re: [PATCH] Convert the VFS cache lock to an rmlock

On Friday, March 13, 2015 06:32:03 AM Mateusz Guzik wrote:
> On Thu, Mar 12, 2015 at 06:13:00PM -0500, Alan Cox wrote:
> > Below is partial results from a profile of a parallel (-j7) "buildworld" on
> > a 6-core machine that I did after the introduction of pmap_advise, so this
> > is not a new profile.  The results are sorted by total waiting time and
> > only the top 20 entries are listed.
> > 
> 
> Well, I ran stuff on lynx2 in the zoo on fresh -head with debugging
> disabled (MALLOC_PRODUCTION included) and got quite different results.
> 
> The machine is  Intel(R) Xeon(R) CPU E5-2680 v2 _at_ 2.80GHz
>  2 package(s) x 10 core(s) x 2 SMT threads
> 
> 32GB of ram
> 
> Stuff was built in a chroot with world hosted on zfs.
> 
> >      max  wait_max       total  wait_total       count    avg wait_avg
> > cnt_hold cnt_lock name
> > 
> >     1027    208500    16292932  1658585700     5297163      3    313  0
> > 3313855 kern/vfs_cache.c:629 (rw:Name Cache)
> > 
> >   208564    186514 19080891106  1129189627   355575930     53      3  0
> > 1323051 kern/vfs_subr.c:2099 (lockmgr:ufs)
> > 
> >   169241    148057   193721142   419075449    13819553     14     30  0
> > 110089 kern/vfs_subr.c:2210 (lockmgr:ufs)
> > 
> >   187092    191775  1923061952   257319238   328416784      5      0  0
> > 5106537 kern/vfs_cache.c:488 (rw:Name Cache)
> > 
> 
> make -j 12 buildworld on freshly booted system (i.e. the most namecache insertions):
> 
>       32       292     3042019    33400306     8419725      0      3  0 2578026 kern/sys_pipe.c:1438 (sleep mutex:pipe mutex)
>   170608    152572   642385744    27054977   202605015      3      0  0 1306662 kern/vfs_subr.c:2176 (lockmgr:zfs)

You are using ZFS, Alan was using UFS.  It would not surprise me that those
would perform quite differently, and it would not surprise me that UFS is
more efficient in terms of its interactions with the VM.

-- 
John Baldwin