Where’s EIOW Heading?

The Exa-scale I/O Workgroup (EIOW – see EIOW.org) accomplished much of what it aimed for this year.  We held requirement gathering meetings all over the world, and did an early decomposition of the architecture.  A new white paper summarizes the present situation, and frankly, it’s more interesting and quite different from what I thought it would be.   I’m particularly pleased that storage experts who have been following this have been scratching their heads about a few requirements – such as the implications of lazy commit and rollback.   To be somewhat bold and Jobs-ian: what the group p

Xyratex Presenting at LAD‘12

This week, on September 24 and 25, the EOFS and OpenSFS are holding the second European Lustre Workshop, now called the Lustre Administrators and Developers Workshop (LAD) . Last year, this workshop was very well attended in its first iteration and this year will provide another great opportunity for global Lustre(R) brethren, from developers to administrators to solution providers, to meet and hear the latest in best practices, developments, as well as interact on real world applications of Lustre.

Heading to SBAC-PAD-2001 in Brazil . . .

This Thursday, I’ll be speaking at South America’s largest HPC conference about breaking the I/O bottleneck in HPC storage so please join us if you’re planning on to be at the conference. I feel that as an industry, we face many architectural challenges around creating a compute/storage architecture that can perform and scale to meet the requirements of tomorrow’s large Petascale and eventually, Exascale systems.

The Exascale Storage Forum at the 2011 IEEE Cluster Meeting

I was recently on a panel discussing the status of exascale storage today.   The discussion was aligned on the 10PF, 100PF and 1EF boundaries.   For 10PF, there's little to say, Lustre is running on 9 of the top 10 systems, including the 8PF Riken system. Much refinement is possible, Fujitsu has done that for Riken (but has maybe not yet released its patches to the general public), and others will undoubtedly do more.