EML: An I/O scheduling algorithm in large-scale-application environments

Distributed file systems have been widely used in many applications to provide high performance. However, large amounts of data-intensive applications often access the data server concurrently, the average completion time is enlarged due to the long request queue on data servers, especially when requests of applications cause a series of synchronous I/O requests. This paper proposes an I/O scheduling algorithm, called EML (equal-length multi-level algorithm) to solve this problem, it can reduce the average response time significantly. We demonstrate the performance improvement versus multi-level queue through both theoretical and experimental analysis. The experiments show that EML algorithm can effectively reduce the average completion time by 30% in 64 concurrent write applications and 50% in 64 concurrent read applications.