arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2202.06744cs.DC
收藏

多核环境下的开销管理

Overhead Management in Multi-Core Environment

Urmila Shrawankar, Mayuri Joshi

摘要:

在多核系统中,诸如进程间通信、依赖性、资源共享和调度、并行化水平、同步、可用核的数量等各种因素影响可能的高性能计算并行化的程度。这些参数如果没有管理到根级别,以后会在执行期间作为开销出现。本文着重讨论了这些并行性参数及其并行化开销,并对它们进行了有效的管理,以便在任意域下实现最优的并行执行。作为一个整体,我们专注于稠密线性代数(DLA)域,特别是矩阵乘法和排序域。这些领域的选择,因为他们发现在科学和数学应用的各个部门的应用。比较分析所获得的结果澄清之间的权衡串行和并行执行DLA的问题,表面的开销和他们可能的有效管理。

英文摘要:

In multi-core systems, various factors like inter-process communication, dependency, resource sharing and scheduling, level of parallelism, synchronization, number of available cores etc. influence the extent of possible High Performance Computing parallelization. These parameters if not managed to the root level, later surface as overheads during execution. This paper emphasizes on these parameters of parallelism, their overheads of parallelization and its effective management for optimal parallel execution under any domain. As a whole, we focus on the Dense Linear Algebra (DLA) domain and specifically on Matrix Multiplication and sorting domains. These domains are chosen as they find application in various sectors of scientific and mathematical applications. The comparative analysis of results obtained clarifies the trade-off between serial and parallel execution of DLA problems the surfacing overheads and their possible and effective management.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑