Priority conscious transaction routing in a real-time shared disks cluster

A great deal of research indicates that the shared disks (SD) cluster is suitable to high performance transaction processing. However, the aggregation of SD cluster with real-time processing has not been investigated. By adopting cluster technology, the real-time services will be highly available and can exploit inter-node parallelism. In this paper, we propose a priority conscious transaction routing algorithm for a real-time SD cluster, which allocates real-time transactions to a node in the SD cluster. Unlike traditional routing algorithms that consider transaction affinity and load balancing only, our algorithm also considers transaction priorities inherent to real-time applications. We evaluate the performance of our algorithm under a wide variety of real-time workloads. The experiment results show that our algorithm outperforms both pure priority-based algorithms and pure affinity-based algorithms.