Escaping the disk bottleneck in fast transaction processing

An approach to avoiding the disk access bottleneck for transaction processing applications is described. The approach is based on the use of primary copy replication, and takes advantage of recent hardware advances such as inexpensive high-speed CPUs and networks, and uninterruptable power supplies. In addition to improving response time, the architecture increases overall availability and reliability. The work described is in a preliminary stage. The overall architecture and its motivation are discussed. File system techniques in the context of transaction processing are reported.<
>