Abstract:
In many scientific Clouds, storing very large amounts of application data remains a great challenge. To provide necessary storage and performance support, one strategy is to distribute data over multiple disks using RAID technologies which are widely available and very robust. Adding new storage disks to cope with large amounts of Application data, requires proper parallel data redistribution techniques to maintain the performance of entire system. In this paper we describe various techniques and algorithms that take advantage from redistribution strategies aiming on increasing the performance of a scalable parallel disk array. We will summarize several recent methods and approaches like SCADDAR, SLAS, ALV and FastScale. We will describe FastScale implementation and propose an algorithm to take in account parity block position structures to enable parallel read/writes on the extended volumes. Numerical results show that FastScale outperforms SLAS under the same workloads. We conclude with a discussion of the expected performance or proposed algorithm and future works on performance evaluation.
