Stroll: A Universal Filesystem-Based Interface for Seamless Task Deployment in Grid Computing

摘要

Developing applications for solving compute intensive problems is not trivial. Despite availability of a range of Grid computing platforms, domain specialists and scientists only rarely take advantage of these computing facilities. One reason for this is the complexity of Grid computing, and the need to learn a new programming environment to interact with the Grid. Typically, only a few programming languages are supported, and often scientists use special-purpose languages that are not supported by most Grid platforms. Moreover, users cannot easily deploy their compute tasks to multiple Grid platforms without rewriting their program to use different task submission interfaces. In this paper we present STROLL, a universal filesystem-based interface for seamless task submission to one or more Grid facilities. Users interact with the Grid through simple read and write filesystem commands. STROLL allows all categories of users to submit and manage compute tasks both manually, and from within their programs, which may be written in any language. STROLL has been implemented on Windows and Linux, and we demonstrate that we can submit the same compute tasks to both Condor and Unicore clusters. Our evaluation shows the overhead of STROLL to negligible. Comparing the code complexity of a STROLL compute task with command-line clients and Grid APIs show that STROLL can eliminated up to 95% of the complexity.