About number of effectively performed streams in multi-core computers with shared memory

Abstract

The article investigates the dependence of the number of effectively executed programs in multi-core computers with shared memory on the parameters of programs and computers. All computer cores execute parallel streams of a single program developed in accordance with the OpenMP API. There are no interactions between streams program. Conflicts can occur only when cores are accessed into shared memory. Suggested the necessary models. Analytical expressions are obtained for the dependence of the number of effectively executed programs on the properties of programs, core and shared memory parameters. The main reason for limiting the number of effectively executed programs is the overload of the shared memory. The threshold value of the number of effectively executed programs has been determined.