A highly reliable and parallelizable data distribution scheme for data grids

摘要

The major drawback of the replication-based RAID (redundant arrays of independent disks) architectures is that, in spite of the high redundancy level, they cannot balance the load increased by the disk failures and this results in reliability and access bandwidth reduction for processing the data access requests. Furthermore, these schemes are not able to determine the actual position of the occurring data block errors (or in-error data blocks). In this paper, to alleviate the addressed problems, we propose a new parity-based striped mirroring scheme called PSM-RA1D in which the striped data blocks are replicated among the other disks, and a parity block is then associated with each stripe. In this method, a data mirroring scheme improves the reliability of the array by directing the read requests to the mirrored copy of data, and the parity data enables the controller to determine the actual position of the in-error blocks. The specific data distribution algorithm proposed in this paper also improves the access bandwidth of the array to serve many more disk requests. The proposed method is compared with similar architectures and the simulation results show that the proposed model due to the data striping scheme is able to provide a significantly higher parallelism between the disk requests, as well as a higher reliability due to the block mirroring scheme and the parity blocks.