Features:

PHIL ANDREWS EXPLAINS StorCloud's SHOWCASE OF HPC STORAGE
by Tim Curns, Editor

StorCloud, a new initiative at SC2004 this year, builds a High Performance Computing (HPC) storage capability to showcase HPC storage technologies (topologies, devices, interconnects). StorCloud's "storage on request" area network capability, comprised of state-of-the-art heterogeneous devices and technology, supports researchers and demonstrates high bandwidth applications.

HPCwire recently spoke with StorCloud's Challenge Team chair, Phil Andrews, to find out more about this initiative. Mr. Andrews is the program director for High-End Computing at the San Diego Supercomputer Center (SDSC).

HPCwire: Please describe StorCloud for our readers. What do you hope to accomplish with this initiative?

Phil Andrews: StorCloud is recognition that supercomputing is no longer just Flops; in fact, if it ever was, that was a long time ago. The I/O capabilities provided by an advanced supercomputing center can be more important to users than the purely computational resources, and deserve equal billing in the pantheon. If anything, it's been the inherent complexity of data storage that has prevented anyone from even pretending that it could be typified by single number. Instead of quoting one characteristic, the StorCloud equipment is actually on the show floor, and real applications will be run, giving a true accounting of how well both the hardware and software perform.

HPCwire: How will participants and conference attendees benefit from this? How will it be available for them to try? What resources will be allocated to conference goers?

PA: Most supercomputer applications take a significant amount of time and effort to prepare, and the StorCloud participants have been working with the vendor supplied storage for some time to port their work and prepare for the show. We hope that the lessons learned will be propagated to the conference goers via talks and papers.

HPCwire: How much preparation has gone into kicking StorCloud into gear? What challenges have stood in your way as you gear(ed) up for SC?

PA: There has been a large amount of preparation required: partly because this is the first time for StorCloud, but also because of the sheer size of the effort. Approximately a petabyte of vendor supplied disk will be on the show floor; more than in almost any large machine room! Coordination and staging have been challenges, but there have been exemplary efforts from both those involved in the challenge and the vendors.

HPCwire: This is being created strictly by volunteers, right? How did you get set to become the Chair for the Challenge Team?

PA: Yes, and I'd like to thank the individuals, organizations, and vendors who have provided time and support. I haven't figured out what happened with me yet, but I definitely got one of the easier jobs.

HPCwire: I'm sure it was well deserved! How will StorCloud leverage the SCinet infrastructure? Could you please explain how this will work?

PA: StorCloud will leverage both the SCinet local and wide area networks. Within the conference center, multiple Fibre Channel and GbE links will move data between the StorCloud booth and the application booths, we hope at many Gigabytes per second. SCinet has provided a 30 Gb/s link to the outside world, and we hope that at least some applications will use it to show what can be done with WAN-connected storage.

HPCwire: Describe the Challenge effort for StorCloud? What has your team done to prepare? What will be accomplished or proved? When will the Challenge take place?

PA: The Challenge is a set of approximately 5 applications which will attempt to show what can be done with the excellent storage resources provided. I don't want to steal the thunder of each of the teams involved, so I encourage people to visit their booths and attend the StorCloud proceedings.

HPCwire: How has the increasing need for dependable and large scale high- performance storage changed the overall HPC landscape?

PA: I/O is now a significant part of any procurement: in the old days, the main compute system would be acquired and you would look under the couch cushions to find the spare change for a few disks and tape drives. Now, it's expected that the data part of the acquisition may range between 20 and 50% of the total, and may determine the compute platform.

HPCwire: What do think is the biggest obstacle to overcome in the next year in regards to your field and expertise in high-end computing?

PA: Processor development is now almost completely driven by small system needs; the investment necessary for new fabrications is truly enormous. The two biggest challenges are to either take the cost-effectiveness of the popular processors and make truly balanced and useful large systems, or to create a working business model for a dedicated supercomputing line. Obviously, data capabilities are extremely important in both efforts.

HPCwire: What's on the horizon for SDSC? Do you see any major initiatives taking place this coming year?

PA: We're continuing to both fulfil current user requirements, and predict new ones. We'll be bringing in a single rack Blue Gene/L system this year, specifically configured for data intensive use.

HPCwire: How do you see the United States' HPC efforts in reference to others around the world? I know you were born and studied in England. Are there large differences in regards to HPC priorities?

PA: There are significant cultural differences between the two countries, and several books have been written looking at them, generally in a humorous vein. And of course, there are significant differences within each of them; they just happen over shorter distances in the UK. Almost as many Britishers as Americans think my central Birmingham accent is from Australia!

HPCwire: What brought you out to San Diego from Pittsburgh? How does it feel to be back?

PA: Western Pennsylvania and Southern California are two of the nicest parts of the USA; they're very different, and both unique. I'll know they've met when I see an Amish buggy on the beach!

HPCwire: Not likely! And finally, I read that you're a champion powerlifter. I must ask: What's the most you've ever lifted?!

PA: My best competition lifts are a Bench press of 237.5 Kg (523 lbs) and a Deadlift of 380 Kg (837 lbs). When things get frustrating, I go to the Gym!

HPCwire: Sounds like a good plan! I hope your stay in Pittsburgh doesn't cause too much stress! Thanks so much for talking to us, Phil.

Phil Andrews is chairing the Challenge portion of the StorCloud initiative. Please see http://www.vtksolutions.com/StorCloud/challenge.shtml for more information on this.