A new option for Amazon Web Services has arrived: the raw computing power of supercomputing clusters now widely used in research circles.
The service, called Cluster Compute, is a variation of one of the earliest services Amazon offered, EC2, or Elastic Compute Cloud. Compared with the standard EC2, Cluster Compute offers more processing power and faster network connections among the cluster’s computing nodes for better communications, Amazon said Tuesday. The service retains the same general philosophy, though: customers pay as they go, with more usage incurring more fees.
The cluster service, which is available with Linux and a customer’s own software added into the mix, is best suited to parallel tasks that can be divided into independent pieces that run simultaneously.
The special-purpose version of EC2 is better than the generic, according to Keith Jackson, a Lawrence Berkeley National Laboratory computer scientist. “In our series of comprehensive benchmark tests, we found our HPC (high-performance computing) applications ran 8.5 times faster on Cluster Compute Instances for Amazon EC2 than the previous EC2 instance types,” he said in Amazon’s announcement.
How fast is it? An 880-node cluster reached 41.82 teraflops, or floating-point operations per second, using the Linpack mathematical speed test. By contrast, the 145th-fastest machine on the most recent “Top500” list of the fastest supercomputers reached a sustained speed of 41.88 teraflops. (Amazon didn’t say whether its test was for sustained performance or for the higher but more fleeting peak performance.)
The service is sold on the basis of much smaller nodes than what was used in Amazon’s test: a server with two quad-core Intel X5570 Nehalem Xeon processors. Each such instance costs $1.60 per hour to use. Alternatively, with a payment of $4,290 for one year or $6,590 for three years, the per-hour fee drops to 56 cents.