A Data Assignment Algorithm Based on Non-centralizing Scheduling of Parallel I/O

摘要

Based on the research situation of parallel I/O data scheduling policies, a new data scheduling policy named the dynamic non-centralizing I/O scheduling algorithm (DNC-IO) is proposed. The main idea is to alleviate the bottleneck of parallel I/O data scheduling. The experimental result has shown the Mean Response Time(MRT) of the DNC-IO policy is shortened 5.23 percent relative to the Adaptive Equi-Partition I/O scheduling policy(AEQUI) when the system overloads. At last, the future development of parallel I/O data scheduling is presented.