[cdwg] Lustre 2.3/2.4 update - October 26th 2012

[Jones, Peter A]

Hi there

Now that Lustre 2.3 is GA our focus will switch to Lustre 2.4, but after my rather succinct release announcement on Monday I wanted to make some fuller closing comments on 2.3.

Firstly I wanted to emphasize the role OpenSFS played in the 2.3 release – in addition to the much-appreciated co-sponsoring of the costs of creating this release (which are quite substantial considering the number of individuals involved over the six months+ taken to complete the release) there is also input from the CDWG. The CDWG is ably run by Pam Hamilton of LLNL and – while it is open to anyone to participate – the most regular and active participants throughout the 2.3 release cycle were from Cray, LLNL, ORNL, Whamcloud/Intel and Xyratex.

Furthermore, I wanted to draw attention to the diversity of development organizations who made code contributions to the 2.3 release – Bull, CEA, Cray, DDN, EMC, Fujitsu, LLNL, ORNL, TACC, Ultrascale, UVT, Whamcloud/Intel and Xyratex. I think that this is another positive side-effect of the switch we made to the contribution model at the beginning of last year - in addition to ensuring that it would be impossible for any one organization to alter the license under which Lustre is distributed, I think that it also removes the complexity associated with contributing code (no complicated forms full of legalese to sign) and so encourages contributions.

Finally I wanted to spotlight the ongoing benefit from our usage of the LLNL-hosted Hyperion test cluster – this enables us to test at scales that would not otherwise be available to us and so is invaluable for us to ensure that the Lustre releases will hold up at the kind of scale it is exposed to in production. Hyperion was having infrastructure upgrades throughout the 2.3 release cycle  and yet the administrators continued to be very accommodating in giving us access to perform release testing. This benefit was increased because LLNL gave us permission to use their Simulated Workload (SWL) test suite during our Hyperion testing and this uncovered a number of issues not seen in other testing (even at scale).

So, on behalf of Whamcloud/Intel and indeed the whole Lustre community I would like to warmly thank OpenSFS, the CDWG, the Hyperion administrators and everyone who contributed towards this release.

Turning to Lustre 2.4, a number of landings have been made to master while 2.3 was in release testing – these include the remaining changes necessary to allow Lustre to optionally run on top of ZFS. Lustre 2.4 blockers can be found at http://jira.whamcloud.com/secure/IssueNavigator.jspa?mode=hide&requestId=10292

Regards

Peter