[Twg] WC and FEFS

[Andreas Dilger]

On 2011-11-18, at 2:39 PM, Nathan Rutman wrote:
> http://www.japancorp.net/press_release/24453/whamcloud_and_fujitsu_to_collaborate_on_lustre_development	
> 
> Does this mean we're going to get Fujitsu's FEFS patches incorporated into the trunk?

We definitely hope so.  As yet we haven't seen anything more than the
presentations at LUG and SC (the one just after mine), so I don't know
what all of the changes are, nor how easy/hard they will be to integrate.

Some of the major improvements previously discussed include:
- fair share: an NRS-like bandwidth management enhancement.  Not sure if
              it is per UID or per job, and whether it has both min and max
              bandwidth tunables, nor who/how it is managed
- 10000 OSTs: a number of fixes and optimizations to reduce memory usage
              and the size of RPCs when not needed
- ping throttle: disable client->server pings, and use something IR-like
              to notify clients when a server is down, and a notify servers
              when a client is down using some external health network
- SPARC arch: big endian support for server and clients, large PAGE_SIZE on
              the server
- ext4 fixes: large block size (on SPARC servers)
- max ACLs: increased max ACL count to 8192 (via larger xattrs, I guess)

Cheers, Andreas
--
Andreas Dilger 
Principal Engineer
Whamcloud, Inc.