Dynamic multiple parity (DMP) disk array for serial transaction processing

The performance of today's database systems is usually limited by the speed of their I/O devices. Fast I/O systems can be built from an array of low cost disks working in parallel. This kind of disk architecture is called RAID (Redundant Arrays of Inexpensive Disks). RAID promises improvement over SLED (Single Large Expensive Disks) in performance, reliability, power consumption, and scalability. However, a general fact about RAID is that the "write" operation is difficult to speedup. In this paper, we propose a new RAID architecture, called Dynamic Multiple Parity (DMP) Disk Array, for serial transaction processing database systems. Serial transaction processing database systems include engineering database systems, fully replicated database systems using a completely centralized algorithm and distributed systems using the conservative timestamp ordering algorithm. DMP Disk Array can significantly increase the I/O throughput by incorporating multiple parity disks. Due to the inherent distributed sparing property, DMP Disk Array can provide normal service to the users under single disk failure condition. Delay and maximum throughput analysis on DMP Disk Array is performed. Results show that, for a typical "write" job proportion of 20 percent, DMP Disk Array can provide nearly 20 percent improvement on I/O throughput over that of RAID level 5 when one extra parity disk is used.