Adaptive transaction scheduling for transactional memory systems

Transactional memory systems are expected to enable parallel pro-gramming at lower programming complexity, while delivering im-proved performance over traditional lock-based systems. Nonethe-less, there are certain situations where transactional memory sys-tems could actually perform worse. Transactional memory systems can outperform locks only when the executing workloads contain sufficient parallelism. When the workload lacks inherent paral-lelism, launching excessive transactions can adversely degrade per-formance. These situations are likely to become dominant in future workloads when large-scale transactions are frequently executed. In this paper, we propose a new paradigm called adaptive trans-action scheduling to address this issue. Based on the parallelism feedback from applications, our adaptive transaction scheduler dy-namically dispatches and controls the number of concurrently ex-ecuting transactions. In our case study, we show that our low-cost mechanism not only guarantees that hardware transactional mem-ory systems perform no worse than a single global lock, but also significantly improves performance for both hardware and software transactional memory systems