Re: [PATCH 7/8] knfsd: repcache: faster hash chain probe

[Greg Banks <gnb@sgi.com>]

On Mon, Oct 16, 2006 at 12:34:54PM +1000, Neil Brown wrote:
> On Wednesday October 11, gnb@melbourne.sgi.com wrote:
> > knfsd: Squeeze some small performance improvements out of the duplicate
> > request cache's lookup function by reducing the number of instructions
> > to compare each entry during hash chain probing.
> > 
> > Comparisons such as the RPC protocol number and version which are going
> > to be the same for almost all entries are now done last.  Comparisons
> > such as xid and client IP address which are more likely to differ
> > between entries are done first.  The client IP address comparison
> > is fixed not to compare 8 bytes of zeroes and to do comparisons as
> > 32b quantities.  Finally, a 32b addition for the timestamp is removed
> > from the loop by taking advantage of a loop invariant.
> 
> And yet:
> 
> > +static inline int
> > +compare_sockaddr_in(const struct sockaddr_in *sina, const struct sockaddr_in *sinb)
> > +{
> > +	return ((*(u32 *)&sina->sin_family == *(u32 *)&sinb->sin_family) &&
> > +		(sina->sin_addr.s_addr == sinb->sin_addr.s_addr));
> > +}
> > +
> 
> You compare family and port before address.... I suspect you would
> find the port is the same between clients more often than the address.

Excellent point, I'll fix that.

> Is there a reason?

Not one that I'd want to defend ;-)

Greg.
-- 
Greg Banks, R&D Software Engineer, SGI Australian Software Group.
I don't speak for SGI.

-------------------------------------------------------------------------
Using Tomcat but need to do more? Need to support web services, security?
Get stuff done quickly with pre-integrated technology to make your job easier
Download IBM WebSphere Application Server v.1.0.1 based on Apache Geronimo
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=120709&bid=263057&dat=121642
_______________________________________________
NFS maillist  -  NFS@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/nfs