Date: Sat, 17 Mar 2001 23:57:45 -0800 From: Peter Wemm <peter@netplex.com.au> To: Matt Dillon <dillon@earth.backplane.com>, Julian Elischer <julian@elischer.org>, cvs-committers@FreeBSD.ORG, cvs-all@FreeBSD.ORG Subject: Re: cvs commit: src/sys/nfs nfs_node.c Message-ID: <200103180757.f2I7vjh19781@mobile.wemm.org>
next in thread | raw e-mail | index | archive | help
Peter Wemm wrote:
> Matt Dillon wrote:
> > I'd like to see this in -stable after the freeze is over!
>
> We cannot wait at Yahoo. Other changes in the name caching have made
> this a showstopper and has halted one of our key 2.2.x -> 4.x upgrade process
es
> as faster machines running 4.x cannot handle the load that 2.2.8 did.
> We're running it already for testing to see if that is the only new problem
> that that group of machines turns up..
Actually, this didn't quite come out the way I meant. At work, we do
our own internal releases etc. There probably wouldn't be more than a
handful of machines running an actual FreeBSD -RELEASE of some form.
In it's simplest form, the following works just as well for fixing the
lousy nfs client performance:
Index: nfs_node.c
===================================================================
RCS file: /home/ncvs/src/sys/nfs/nfs_node.c,v
retrieving revision 1.36
diff -c -r1.36 nfs_node.c
*** nfs_node.c 2000/01/05 05:11:36 1.36
--- nfs_node.c 2001/03/18 07:55:45
***************
*** 75,94 ****
/*
* Compute an entry in the NFS hash table structure
*/
! u_long
nfs_hash(fhp, fhsize)
register nfsfh_t *fhp;
int fhsize;
{
register u_char *fhpp;
! register u_long fhsum;
register int i;
fhpp = &fhp->fh_bytes[0];
! fhsum = 0;
! for (i = 0; i < fhsize; i++)
! fhsum += *fhpp++;
! return (fhsum);
}
/*
--- 75,98 ----
/*
* Compute an entry in the NFS hash table structure
*/
! #define FNV_32_PRIME ((u_int32_t) 0x01000193UL)
! #define FNV1_32_INIT ((u_int32_t) 33554467UL)
! u_int32_t
nfs_hash(fhp, fhsize)
register nfsfh_t *fhp;
int fhsize;
{
register u_char *fhpp;
! register u_int32_t hval;
register int i;
fhpp = &fhp->fh_bytes[0];
! hval = FNV1_32_INIT;
! for (i = 0; i < fhsize; i++) {
! hval *= FNV_32_PRIME;
! hval ^= (u_int32_t)*fhpp++;
! }
! return (hval);
}
/*
> > -Matt
> >
> > :Peter Wemm wrote:
> > :>
> > :> peter 2001/03/16 21:43:01 PST
> > :>
> > :> Modified files:
> > :> sys/nfs nfs_node.c
> > :> Log:
> > :> Dramatically improve the **lame** nfs_hash(). This is based on the
> > :> Fowler / Noll / Vo Hash (http://www.isthe.com/chongo/tech/comp/fnv/).
> > :>
> > :> This improves hash coverage a *massive* amount. We were seeing one
> > :> set of machines that were using 0.84% of their 131072 entry nfsnode
> > :> hash buckets with maximum chain lengths of up to ~500 entries. The
> > :> machine was spending nearly 100% of its time in 'system'.
> > :> A test with this has pushed the coverage from a few perCent up to 91%
> > :> utilization with a max chain length of 11.
> > :>
> > :> Submitted by: David Filo
> > :
> > :tell David that it's nice to see he's still keeping his hand in..
> > :
> > :>
> > :> Revision Changes Path
> > :> 1.43 +17 -9 src/sys/nfs/nfs_node.c
> > :
> > :--
> > : __--_|\ Julian Elischer
> > : / \ julian@elischer.org
> > : ( OZ ) World tour 2000-2001
> > :---> X_.---._/
> > : v
> >
> >
>
> 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
>
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?200103180757.f2I7vjh19781>