How Do You Define a Supercomputer ?

By Shai Fultheim – Founder, President and CEO

I’ve been thinking a lot about Supercomputing lately, especially as we prepare for one of the two large “Supercomputing” events that is coming up. The International Supercomputing Conference (known as ISC), in Leipzig, Germany– is next week, and in November is the Supercomputing 13 (known as SC13) show in Denver, Colorado. But really, what is a Supercomputer ?

In the “old” days, a Supercomputer was a monster machine, set apart from the rest of a company’s data processing systems. It was only meant to be touched or administered by specialists, those trained in Supercomputers. Users were granted special access and had to also be trained to develop special applications that would only run on that one computer. But times have changed. The fastest Supercomputers today are really a bunch of smaller servers all working together to solve a problem. The list of the Top 500 supercomputers today is based on a benchmark which no one really runs all day, which is quite easy to program to run across the entire cluster. This isn’t real life. Many algorithms are not easy to spread across this cluster-based “Supercomputer”. Although throwing together bunches of smaller computers to be called a supercomputer is possible, the administration and programming is a challenge.

In my view, “Supercomputer” is all about one computer – with shared memory and I/O subsystems. It is not about “super-large-collection-of-small-machines-that-I-need-to-program-and-administer-individually”. Enter the ability to create a single system (that is the critical element) from a bunch of these smaller servers, eliminating the complexities of a cluster. By using virtualization – or in other words: software to replace propriety hardware – a Supercomputer can be created from smaller pieces. Due to the power of virtualization, developers and users will not have any idea of what is happening under the hood.

Simpler to program, simpler to run, and flexible. You can create a Supercomputer as you need one, no need to fork out to hundreds of little machines to run your code, or shell out millions of dollars for a monster. Create your own!!! Check us out at www.scalemp.com.


Leave a Reply