FastPara: A High-Level Declarative Data-Parallel Programming Framework on Clusters

Y. Mao, Y. Gu, J. Chen, and R.L. Grossman (USA)


Parallel-programming, framework, data-parallel, cluster


This paper presents FastPara, a C++ programming framework and associated runtime support for writing and running data-parallel applications in computer cluster environments. With FastPara, the user writes a declaration of the data to be exchanged between processes for a certain data-parallel processing algorithm. FastPara will handle all the process communications without any users’ efforts. Complex data structures and variable-length arrays are well supported so that users can easily apply this framework to their data-parallel applications. FastPara’s runtime provides built-in support for process management and fault tolerance for the parallel processing. FastPara can greatly simplify the development cycle. Based on the performance tests we did, FastPara demonstrated good communication efficiency and little process coordination overhead. We believe that FastPara is a very useful framework in developing data parallel applications.

Important Links:

Go Back