G'day,
This patch set renovates the knfsd duplicate request reply
cache aka the repcache, primarily aimed at scalability to
large workloads. Substantially identical patches to these
have been shipping for some months as part of SGI's NAS
Server Bundle.
1 minor cleanups
Minor whitespace, formatting and comment changes.
2 make REQHASH a function
Make the REQHASH() macro an inline function and tweak
it to handle observed client xid behaviour.
3 inline hash_refile
Explicitly inline hash_refile(), it's only used in one place.
4 split hash index
Split the repcache hash into multiple buckets on SMP platforms,
to avoid lock contention and bouncing cachelines.
5 dynamically expand under load
Allow the repcache to expand on demand, when too young an
entry falls off the end of an LRU chain.
6 use client IP address in hash
Use the client's IP address in the repcache hash function, to
avoid contention on computer cluster workloads.
7 faster hash chain probe
Squeeze some small performance improvements out of the repcache's
lookup function by speeding comparisons during hash chain probing.
8 add statistics
Add more statistics to the repcache, and reduce contention
by keeping stats per-bucket and aggregating them on demand.
P.S. AFAIK this is the repcache work referred to in Olaf Kirch's
talk at OLS this year.
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 - [email protected]
https://lists.sourceforge.net/lists/listinfo/nfs