Date: Sat, 17 Mar 2001 03:20:19 -0800 From: Peter Wemm <peter@netplex.com.au> To: David Malone <dwmalone@maths.tcd.ie> Cc: cvs-committers@FreeBSD.org, cvs-all@FreeBSD.org Subject: Re: sys/alpha/include types.h sys/i386/include types.h sys/ia64/include types.h sys/powerpc/include types.h sys/kern vfs_cache.c sys/nfs nfs.h nfs_node.c sys/sys fnv_hash.h Message-ID: <200103171120.f2HBKJh08018@mobile.wemm.org> In-Reply-To: <200103171050.aa94864@salmon.maths.tcd.ie>
next in thread | previous in thread | raw e-mail | index | archive | help
David Malone wrote: > > Make the name cache hash as well as the nfsnode hash use it. > > Any information on how this changes the performance of the name > cache? The NFS commit message had some interesting stats - does > this yield similar improvements? This is only minor.. The v_id was already factored into the name cache hash, so there was a fair distribution already. The testing that I did suggested that it smothed things out a little more, but it isn't worth getting too excited about. It was the NFS one that was the real killer. We reran some tests on a machine that had 128K slots in the nfsnode hash. before: utilization = 0.88%; max chain length = ~950 after: utilization > 87%; max chain length = 17 We aborted this test to do some more accurate stats. The percentage utilization of those 128K slots was still gradually increasing and may have got higher still if we had left it to run. There may still be other problems exposed by this data set, so we are trying to finish the tests. Cheers, -Peter -- Peter Wemm - peter@FreeBSD.org; peter@yahoo-inc.com; peter@netplex.com.au "All of this is for nothing if we don't go to the stars" - JMS/B5 To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe cvs-all" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200103171120.f2HBKJh08018>