Amazon’s HPC cloud: supercomputing for the 99%
The Amazon Elastic Compute Cloud is becoming increasingly popular for high-performance computing. It’s now capable of running many of the applications that previously required building out a large HPC cluster or renting time from a supercomputing center. But as you might expect, Amazon EC2 can’t do everything a traditional supercomputer can.
Scientific applications that require super-fast interconnects are still the reserve of the IBM Blue Genes and Cray supercomputers of the world. However, Amazon’s adoption of 10 Gigabit Ethernet connections in its high-end cluster compute instances have expanded the usefulness of EC2.
As we’ve reported before, clusters of 30,000 cores, even 50,000 cores have been run on Amazon’s cloud for real-world scientific applications. A caveat is that these tend to be “embarrassingly parallel” applications, in which the interconnect speed isn’t important because each calculation runs independently of all the others…

