Asynchronous PageRank computation in an interactive multithreading environment

Abstract. Multiple-core processors set the new hardware standard for typical scientific computing platforms thus driving the need for algorithms making extensive use of Thread Level Parallelism (TLP). On the other spectrum end of concurrent platforms, large-scale distributed systems dictate the design of novel algorithms with minimum synchronization constraints and maximum self-adaptation to inevitable dynamic changes in their execution environment. We advocate the asynchronous computation model [1] as a suitable framework for building such algorithms and propose Jylab [2] as an interactive multithreading workbench for comfortably exploring their behaviour. We demonstrate the flexibility of this environment in the asynchronous computation of PageRank and comment on some interesting properties of the computation itself. In this context a multicore machine naturally boosts the performance of an application consisting of asynchronously computing threads and Jylab facilitates such multithreaded designs. Ultimately concurrency at CPU level can drive decisions for alternative ways of organizing concurrency at Internet level but can also serve as 3 4 the final execution platform