چكيده به لاتين
In recent years, many scientific conferences have been held throughout the world, whose results indicate several unanswered questions in different sciences including weather, atomic physics, material science, chemistry, biology, and security. Answering these questions is not possible without developing Exascale computing systems. Exascale computing systems involve many challenges including high energy consumption, scalability, high rate of data transmission, and a high error rate. Exascale computing systems have a high volume of parallelism on the intra-node level, due to having nodes with about 1000 processing cores. This causes an increase in the number of application program threads. Managing the creation and termination, and the synchronization of these threads is a difficult task. In addition, application programs emerging in High Performance Computing systems, have composition and multi-enclave architectures. This new architecture of application programs, high rate of parallelism, and heterogeneous architecture of processing nodes, necessitate the operating system and runtime systems to consider these challenges when scheduling the intra-node processes. Our goal is to create application programs on processing cores in a Exascale computing system node, such that some of the limitations considered are realized. The main idea is to use binary integer linear programming and binary quadratic programming, so that we can decrease the data transmission rate on an NoC level that reduce energy consumption and system process turnaround time. Results of the experiment show the superiority of binary quadratic programming method compared to binary integer linear programming in scheduling the processes by reducing the number of variables.
Keywords: High performance computing, Exascale computing systems, Scheduling, optimization.
