Gordon the supercomputer is intense about data
300TB of flash = Big Data, fast
Posted in HPC , 23rd November 2011 23:00 GMT
Watch Now : Virtual Machine Movement with Hyper-V
SC11 According to San Diego Supercomputing Center chief Mike Norman, his brainchild 'Gordon' is the world’s first data intensive supercomputer.
In the works for two years, Gordon was being shipped from system house Appro to its new home in San Diego last week during SC11 in Seattle. In the video, I catch up with Mike and discuss what he means by a ‘data intensive supercomputer’ and how Gordon is different from what’s come before.
One of the biggest differences, of course, is the whopping 300TB of Intel MLC flash storage feeding Gordon’s 1,024 compute nodes. We also talk about the advantages/disadvantages of flash vs. spinning disk and, of course, kick around the cost issue.
Does the performance advantage of flash outweigh the significant cost premium? Steve Lyness, HPC Solutions Engineering VP from Appro (the folks who built Gordon), talks about how the use of flash in Gordon reduces the ‘dollars per science’ ratio by radically reducing the amount of time CPUs sit idle while data is transferred from and to storage. ®
Send corrections
Data control in the cloud
The new Office Garage series:
Enabling efficient data center monitoring
IT infrastructure monitoring strategies
