In the area of cluster computing, single system image(SSI) services is an approach to get higher utilization of various computing resources. In such services, Single Disk I/O(SDIO) is an approach to provide SSI service at the disk level. SDIO service is achieved by the cooperation of device driver in each cluster node.
At the previous implementation of SDIO driver module, we could achieve a virtual disk storage through many disks attached in each cluster node. But it shows poor I/O performance due to the lack of parallel disk I/O capability.
In this thesis, we newly implement SDIO driver module to enhance the I/O performance by supporting for efficient SDIO service. When using distributed disks, network operation is inevitable but it pure overhead. To reduce the number of network operations, we use Request Grouping and Pattern-based Data Pushing. By Request Grouping, we also reduce the number of disk I/O operations at remote nodes.
According to experimental results, we could obtain performance gain by parallel I/O mechanism and Request Grouping. In the case of Pattern-based Data Pushing, experimental results show that miss penalty for pushed data is very high, so there needs more precise predicting algorithm