Improving the Performance of Data Servers Using Array Codes

Abstract

This paper discusses improving performance (throughput) of data server systems by introducing proper data redundancy into the system. General performance properties of a server system with redundant data are described. We show that proper data redundancy in a server system can significantly improve the performance, in addition to the reliability of the system. Two problems related to the performance together with their solutions are proposed, namely, the problems of efficient data distribution scheme for the severs and data acquisition scheme for the client. Both schemes utilize array codes, a class of error-correcting codes whose encoding and decoding procedures only use simple binary exclusive-OR operations, which can be implemented efficiently in software and/or hardware. Construction of general MDS array codes suitable for the both schemes is discussed. A new property of MDS array codes, called the strong MDS property, is also defined to improve the data acquisition performance. A method for modeling data server performance and the related experimental results are presented as well.