Date: Tue, 10 Nov 1998 12:48:55 -0500 (EST) From: Alfred Perlstein <bright@hotjobs.com> To: Mike Smith <mike@smith.net.au> Cc: current@FreeBSD.ORG Subject: Re: Simple NFS ACCESS caching, call for testers Message-ID: <Pine.BSF.4.05.9811101214590.370-100000@porkfriedrice.ny.genx.net> In-Reply-To: <199811101002.CAA01721@dingo.cdrom.com>
next in thread | previous in thread | raw e-mail | index | archive | help
vfs.nfs.async: 1 vfs.nfs.gatherdelay: 10000 vfs.nfs.gatherdelay_v3: 10000 vfs.nfs.defect: 0 vfs.nfs.diskless_valid: 0 vfs.nfs.diskless_rootpath: vfs.nfs.diskless_swappath: vfs.nfs.access_cache_timeout: 6 vfs.nfs.access_cache_hits: 225001 vfs.nfs.access_cache_fills: 10981 nfsserver:/xxx /xxx nfs rw,tcp,bg,nfsv3,-r32768,-w32768,intr 0 0 Those are the mount options, I didn't notice a substantial increase in performance. However FreeBSD was about 30-50% faster untarring files from NFS->NFS*. When doing parallel "ls -lR" on a large+deep NFS tree linux beat freebsd by about 1 second. (*) I suspect it would have been faster had I not been running x11amp (idle), for some reason it really kills NFS access. I'm trying to figure out some sorta test that would stress the cache but couldn't think of anything besides the two I just ran. Alfred Perlstein - Programmer, HotJobs Inc. - www.hotjobs.com -- There are operating systems, and then there's FreeBSD. -- http://www.freebsd.org/ 3.0-current On Tue, 10 Nov 1998, Mike Smith wrote: > > The attached patch adds a trivial cache for NFS ACCESS operations, > which may provide a moderate to substantial performance improvement in > some cases. > > The issues surrounding caching these requests are actually quite > subtle, and it's not immediately clear that a more sophisticated > approach would actually yield great results in many more cases than > this trivial one does. The trivial implementation has the advantage of > simplicity. 8) > > If you have an NFS v3 server that you beat heavily on, I'd love to know > whether these changes make any difference to you. > > Apply the patches to a -current kernel (they should apply fairly > cleanly to a -stable kernel as well, but I haven't tried this). > > The new kernel has three new sysctl options: > > vfs.nfs.access_cache_timeout > > The time (in seconds) for which an ACCESS result is cached. > Try values from 2 to 10 or so. A value of 0 (the default) > disables caching. > > vfs.nfs.access_cache_hits > > The number of access calls that have been satisfied from > cached entries rather than wire calls. > > vfs.nfs.access_cache_fills > > The number of access calls that have had to go to the wire > to be satisfied. > > Trivial testing tends to indicate that operations involving a single > UID and a large directory hierarchy may benefit substantially from > this, but I really need more results before I can commit. > > > To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?Pine.BSF.4.05.9811101214590.370-100000>