Parallel biocomputing

Here we briefly describe how to use commodity hardware to build a low-cost, high-performance compute cluster, and provide an in-depth example and sample code for parallel execution of R jobs using MOSIX, a mature extension of the Linux kernel for parallel computing. A similar process can be used with other cluster platform software.As a statistical genetics example, we use our cluster to run a simulated eQTL experiment. Because eQTL is computationally intensive, and is conceptually easy to parallelize, like many statistics/genetics applications, parallel execution with MOSIX gives a linear speedup in analysis time with little additional effort.We have used MOSIX to run a wide variety of software programs in parallel with good results. The limitations and benefits of using MOSIX are discussed and compared to other platforms.With the widespread use of high-throughput genomic technologies, there is currently a great interest in running these computations in parallel. The last few years have seen a sharp increase in the number of both freely and commercially available packages for parallel computing. For a good overview of parallel computing in R see [1] and the R task page for parallel computing [2]. These packages generally run across a single CPU with multiple cores [3], multiple CPUs or cores (e.g., snowfall and sfCluster [4]), and there is also progress on using graphics processors for parallelization with gputools [5]. Some of these try to automate the parallelization process, but generally still require some setup, and require no modification to shared variables. They also require you to have cluster software installed, such as LSF [6], or Rocks [7], to name a few. Here we report good success on building and running a cluster with a more universal and mature tool, MOSIX [8]. MOSIX is a linux kernel patch that, once compiled and installed on a cluster, provides single system image (SSI) parallel computation.The steps required by the user to use MOSIX are conceptua