Supercomputing Takes to the Cloud


The promise of cloud computing has come not only to web startups seeking cheap storage for photos or a way to handle a viral hit without owning a data center, but to big industry, thanks to Tata and folks using Amazon to offer supercomputing as a service. John West over at HPCWire pointed me to this story about Computational Research Laboratories, a Tata Sons’ subsidiary that’s offering its Eka supercomputer through the cloud to companies such as Boeing.

Earlier this month, Amazon (s AMZN) unveiled MapReduce for its cloud, which provides a supercomputer-like ability to manage and thus crunch large data sets in Amazon Web Services. Other researchers are offering their data and software for supercomputing jobs via Amazon’s EC2 as well. So while folks may consider cloud computing good for startups and side projects in the enterprise, scientists and R&D groups are taking the cloud for a spin. Given that trends in supercomputing hit the mainstream IT industry sooner or later, efforts made here will be playing out in the corporate world sooner than we think.



To be fair, super computing has been about parallel, distributed computing for a while. (Think of SETI@home as an example) The cloud is (almost) merely giving it a contemporary name. Talking of Map/Reduce, and keeping semantic jugglery aside, one can also ask if Google (search) is not a massive super computer?

However, semantics are worth considering. So, the question that begs to be asked is – what exactly do we mean by the cloud? Is it predominantly webscale computing infrastructure – or data stores – or software frameworks – or a combination? or something else?

As a side note – what one does find intriguing is how jargon takes over mind space. Cloud is clearly in fashion. And we find several attempts to re-label everything “internet” as “cloud”. I can think of a few headlines – TV takes to the cloud (think Hulu, Vudu). Phones in the cloud (skype), Cloud-sourcing, recipes in the cloud,…

Stacey Higginbotham

Vinay, you’re correct about jargon and paying attention to semantics. Let’s call it supercomputing as a service. There are two points I was interested in. Subscription access to an actual supercomputer via the web (cloud) and the idea of performing HPC jobs in an x86 compute cloud such as Ec2.

John West

Stacey – you are quite right in catching the distinction; it’s a distinction that many who aren’t steeped in the old line tradition of big iron HPC miss.

I think one of the things that will be very interesting to watch is the degree to which HPC applications and traditional scale out cloud infrastructure will influence each other in the coming months. It is not yet clear whether scientists will attempt to package or refactor more traditional applications (which require high bandwidth, closely-coupled communications in many important domains) in ways that are well-suited for the cloud infrastructure, or whether cloud infrastructure providers will offer options that provide hardware suited for them. I think if either happens, the former is more likely than the latter, and that could be the catalyst for a lot of interesting innovation in scientific computing.

I’d encourage you to keep an eye on as well as HPCwire; insideHPC is my full time outlet for news related to the HPC ecosystem.


A friend of mine mentioned Convey Computers (ex Convex team, Richardson TX) that has built a really cool x86 based compute-block for specialized clouds. The spex look pretty good and may be better suited to ‘super’computing tasks than plain jane commodity x86 server based clouds.

Comments are closed.