Applying Batch Processing to Improve Performance of iSCSI Storage System

iSCSI storage systems follow SCSI protocol over IP network, but the transport of large number of small I/O such as SCSI commands, status, and tiny data over network seriously affects the performance of storage systems. The conventional method of batch processing combines several small I/O into a large one. However, this method is lack of quantitative analysis and the value of parameter K is set based on experience. Firstly the mathematical model of SCSI commands batch processing is created through queuing theory. Then the software module called K-batch is designed in iSCSI storage protocol architecture. Finally the iSCSI storage system with K-batch is tested under certain network scenarios built on ns-2. In the experiments, as the command arrival rate changes, the K-batch real-timely adjusts the value of parameter K correspondingly. The results show that the mean response time of SCSI commands is significantly reduced, and so the batch processing strategy can improve the performance of iSCSI storage system.