Performance Evaluation of MPI, UPC and OpenMP on Multicore Architectures

Abstract

The current trend to multicore architectures underscores the need of parallelism. While new languages and alternatives for supporting more efficiently these systems are proposed, MPI faces this new challenge. Therefore, up-to-date performance evaluations of current options for programming multicore systems are needed. This paper evaluates MPI performance against Unified Parallel C (UPC) and OpenMP on multicore architectures. From the analysis of the results, it can be concluded that MPI is generally the best choice on multicore systems with both shared and hybrid shared/distributed memory, as it takes the highest advantage of data locality, the key factor for performance in these systems. Regarding UPC, although it exploits efficiently the data layout in memory, it suffers from remote shared memory accesses, whereas OpenMP usually lacks efficient data locality support and is restricted to shared memory systems, which limits its scalability.

A Middleware for Concurrent Programming in MPI Applications

“…If shared memory is available on a particular hardware platform, notably multi-core processes, an MPI implementation may use it to provide effective interprocess communication for processes located on the cores of an individual CPU. While this is effective enough to compete head-on with OpenMP and UPC [7], the sole purpose of this optimization is to make efficient use of contemporary CPU designs. The primary unit of processing is and remains that of a process.…”

Transitioning Data Flow-Based Visualization Software to Multi-core Hybrid Parallelism

“…Early hybrid parallelism work focused on benchmarking well-known computational kernels [5], [6]. Subsequent work explored specific impacts of hybrid parallelism for visualization [7].…”

Hybrid Parallelism for Volume Rendering on Large-, Multi-, and Many-Core Systems

“…The hybrid model allows data movement among nodes using traditional MPI motifs like scatter and gather, but within nodes using shared-memory parallelism via threaded frameworks like POSIX threads or OpenMP. Previous work comparing distributed-memory versus hybridmemory implementations (e.g., [21], [22]) has focused on benchmarking well-known computational kernels. In contrast, our study examines this space from the perspective of visualization algorithms.…”