I'm back from the annual Supercomputing (SC) conference in Tampa. As always there was a lot of cool stuff going on: despite the name, this is just a great place to go to see innovation in technology and its applications. A few things that impressed me:
- OSU's Introduce IDE for Globus Web Services (see picture) being used to create and deploy new services in a few minutes. All those creating services manually should immediately switch to using Introduce!
- The Netezza storage appliance.
As I understand it, they place FPGAs right next to the disks and thus
can process SQL queries 10-100 times faster than conventional
databases. LLNL used a Netezza system and 40 lines of SQL to run a graph analysis problem 10x larger than one that took 5000 CPUs (and much complex programing) the year before.
- The work of Canadian teams in Victoria, WestGrid, and elsewhere, who have been doing great things with Globus, including GT4, for many years.
We participated in the HPC Profile Interoperability Event, in which multiple groups ran execution services compliant with OGF specifications such as BES and JSDL. Thanks to Peter Lane for making that happen on our side. Andrew Grimshaw described this as the "Great Leap Forward." Given that two years of work on BES has produced a spec that covers only a fraction of what is in the most widely used execution service, GRAM, and in particular does not address key issues such as file staging and authentication, that claims seems overstated. Certainly the specification allows companies to claim that they are "HPC Profile compliant," but so far it does little for interoperability. We should continue to push for true standardization.