Introducing Windows HPC Server
Tuesday, July 19, 2011
High-performance computing has traditionally been its own world. Applications are "jobs", programs are "codes", and people refer to problems as "embarrassingly parallel" without chuckling. Part of the reason for this insularity is that not too many people worked in this world--most organizations couldn't afford the big compute clusters it required--and so as with any subculture, a distinct language and world view emerged.
Today, however, public cloud platforms make important aspects of HPC accessible to everybody. This is definitely a good thing. Yet using these new capabilities requires understanding the space. What kinds of problems are right for clusters? When does it make sense to use HPC technologies, and how do those technologies work? And, of course, what exactly does it mean for a problem to be embarrassingly parallel?
To help answer these questions for people who work with Windows, I've written a Microsoft-sponsored introduction to Windows HPC Server. The paper's goal is to provide an up-to-date overview of this technology, including its support for HPC in the cloud. The target audience is pretty much anybody: It doesn't assume previous knowledge of cluster computing or HPC in general. If this sounds interesting, the paper is available here