Exploring Object-Level Parallelism on Chip Multi-processors

摘要

Object-oriented programming languages are prevalent at present and the inherent parallelism of a wide range of applications can be expressed as groups of concurrent objects. This paper proposes a new object-level parallel programming model to meet the challenge of parallel programming on chip multi-processors. This object-based computation model maps multiple software objects onto different physical processor cores at runtime. Independent objects can run concurrently and communicate with each other by synchronous, asynchronous, future and broadcast messages. In order to verify the efficiency of this model, a library implementation CPPLib of this model based on C++ is presented. The library implementation of this model requires no extension to existing object-oriented programming languages and provides good salability and portability. The performance of CPPLib is evaluated on both a dual-core desktop computer and an 8-core server using standard benchmarks. Experimental results show that CPPLib achieves average speedup levels of 1.5 and 3.6 on the dual-core desktop computer and the 8-core server, respectively, compared with their serial version.