Impact of workload partitionability on the performance of coupling architectures for transaction processing

The authors present an analytical study on the robustness (in terms of performance) of three coupling architectures for transaction processing, namely, shared nothing (SN), shared disk (SD), and shared intermediate memory (SIM), where a shared intermediate level of memory, which is at the next level to the main memory in the storage hierarchy, is introduced. Affinity clustering of workload, which attempts to partition the transactions into affinity clusters according to their database reference patterns, can be used to reduce the coupling degradation under the different architectures. However, partitioning the workload based on affinity and at the same time balancing the load in each cluster is a difficult task, and sometimes cannot be achieved. The authors investigate the impact of affinity clustering on the performance of these three different architectures under various types of workloads