The cycle server: a Web platform for running parallel Monte Carlo applications on a heterogeneous Condor pool of workstations

摘要

We have developed a scientific computing tool which we call the cycle server This tool enables Monte Carlo computations on the Condor cycle scavenging system. The tool presents users with a graphical Web interface which hides the complexity of distributed computing from users. Users need only upload codes and download results via the Web GUI anywhere there is Web access. Parallel random number support (via SPRNG random number library), running tasks on heterogeneous platforms (via the remote compiler), and using Condor to migrate jobs from busy or dead machines to idle ones, are all integrated seamlessly with the tool and are transparent to the user. At the same time, the cycle server acts as a gate-keeper for the workstation pool. No user accounts are needed on pool computers, and jobs only run at otherwise idle times without accessing local file systems. We show the utility of the cycle server on a large Monte Carlo computation from biochemical physics.