Task scheduling in the parallel processing of STAP

Abstract

The traditional parallel processing methods of STAP (Space-Time Adaptive Processing) schedules the algorithm to different processors of specific hardware system based on coarse-grained task division to improve the calculation throughput of the system by pipeline processing between processors. But there are two disadvantages. Firstly, coral-granularity division hinders the parallelism of the algorithm. Secondly, the traditional processing method only takes affects on specific hardware system. This paper puts forward a new parallel processing method based on fine-grained task scheduling, which consists of three steps as follows: Establishing fine-grained task model of STAP algorithm in the form of DAG (Direct Acyclic Graph); Describing different target hardware systems by uniform topology model; Scheduling task model to processors in the topology model in fine-grained task manner. The experiment result shows that the parallel method achieves a favorable speedup, and more flexible adaptation to different STAP applications. © 2012 Springer-Verlag.