Visual Modeling for Multi-level Parallel Computing Environment Based on DSL

As the increasingly high demands of computer processing power and speed, parallel computing has become an inevitable trend. With the development of technologies such as clusters and multi-core processors, there has achieved a very good support for parallel computing at the hardware level. But the inherent difficulty of parallel programs and programmers' habitual thinking for a serial program severely limits the popularization of parallel programs in a variety of areas of high-performance applications. In the system of PAVM, we have achieved the visual modeling of single-level parallel application, as well as the corresponding model verification and code generation work. Based on PAVM, this paper proposes a visual modeling methodology based on the hierarchical idea. A visual modeling system called M-PAVM is designed and implemented which supports three levels modeling for multi-core cluster environments. The hierarchical modeling scheme designed in this paper divides the process of parallel application modeling into three levels: task configuration, process task assignment and algorithm implementation. The first level is responsible for coarse-grained task division and for the mapping from tasks to computing nodes. The second level is used for designing the division, distribution and communication of subtasks, which is aimed at getting well-defined threads. And the last level is used for the design of algorithm and control logic in tasks, which is unrelated to the platform architecture. The application of gene matching in bioinformatics is presented to verify the system's effectiveness.