SC08: Day wrap-up

I saw many things at SC08 … first off, most of the people we saw running disks were running some sort of multi-pipe direct attached storage with RAID0’s. Yeah, this shows bandwidth real well. Not how users really run them, but it shows some nice inflated numbers. Compare this with a RAID10 running over a single iSCSI 10 GbE connection. Most folks are used to slow iSCSI, and can’t believe our numbers, until they see them.

Second, I noticed something about the traffic/attendance. Much reduced over last year. Ok, Reno was a tighter packing, less space, so greater density. But even (mentally) correcting for the difference, it looks to be a substantially smaller crowd here. Maybe this is me … I am not sure. Third, some of the recent high flyers aren’t there, some for obvious reasons, some less so. Linux Networx, which (famously) imploded in January isn’t there apart from distant echos in the ICE product at SGI. I did not detect a “ClearSpeed” booth. Fourth, accelerators are all the rage. Some are doing CUDA/Tesla variants, few are doing ATI, and some diehard FPGA types are out there. One company seemed to have a clue on the FPGA side, basically the pricing for the boards with FPGAs were reasonable. Understand that I found this surprising. The economics of FPGA accelerators have been badly skewed … which IMO served to exclude FPGAs from serious considerations for many users. With the rise in GPU usage for computing, some of the FPGA folks finally started to grasp that you can’t charge 10x node price for 10x single core performance. Understanding that you need 10x node performance (running all 8-16 cores in a node) for an accelerator to be meaningful at about 1x node price at worst. The vast majority of FPGA computing groups don’t grasp this, pointing instead to technological reasons why they are better than GPUs. Also the programmability, and portability … writing code in a high level and being able to move the code binary to similar cards … are major issues, one GPUs have well solved, and something FPGAs do not have solved. Fifth, reception to the concept of ΔV, and the usual reaction we are getting to customers seeing it, is quite striking. People seem to like the idea of something designed to enable them to spend less, yet do more.