Calculation models and task scheduling of parallel heterogeneous environment
|School||Qufu Normal University|
|Course||Applied Computer Technology|
|Keywords||Heterogeneous Computing MPI Parallel Computing Model Task Scheduling|
The lower cost of computer hardware, rapid promotion of microprocessor processing power and high-speed development of network technology have established the conditions for parallel computing using common computers. Generally, this kind of parallel computing systems has a heterogeneous nature. Heterogeneous parallel computing systems have been widely used in fields of science and commerce. Research on heterogeneous computing systems has become a research focus in recent years. Therefore, the study of task scheduling and model of parallel computing in heterogeneous environments is very important.First, this paper introduces the concept of parallel computing and the classification of parallel computing systems, then it analyzes the heterogeneous parallel computing systems and discusses the parallel computing models put already forward to this day. Parallel computation model plays a role as a bridge between hardware and software in parallel computing system. A variety of characteristics of parallel computing system are abstracted to obtain the parallel computing model. For a parallel application, algorithms are designed and analyzed on the parallel computing model and carried out by hardware through compiled high-level language. However, with the development of parallel computing, there is no general-purpose computing model. Parallel computing models which have been proposed either are simple or too abstract to apply to specific circumstances. Basing on the BSP model, this paper presents a non-exclusive heterogeneous asynchronous parallel computing model—NHA-BSP model. Theoretical analysis shows that the NHA-BSP model describes the system performance parameters more accurately and depicts non-exclusive and heterogeneous characters of parallel computing systems and their impacts to parallel computing implementation efficiency in detail. Moreover, the model increases system throughput and improves the implementation efficiency of parallel program through allowing asynchronous implementation of the program. The experiment results verify the availability of the model and the paper gives the method of program optimization.Heterogeneous parallel computing systems consist of multiple heterogeneous processors and processors are connected through different communication links. Parallel task scheduling algorithm plays an important role in the computational efficiency of the whole system. Therefore, the optimal scheduling strategy is another research content of this article. This paper describes content related task scheduling, including the four steps of heterogeneous computing as well as the classification of task scheduling, and focuses on the heuristic scheduling policy of static task scheduling. Based on Critical Path On a Processors CPOP algorithm under Heterogeneous parallel computing environment, this article proposes a scheduling algorithms based task replication. This algorithm has the same time complexity as CPOP algorithm. Through case analysis and simulation experiment, it can be seen that this algorithm achieves higher efficiency than the CPOP algorithm.