GPU Acceleration in OLTP Execution

This lecture discusses the use of GPU to accelerate OLTP execution, presenting GPU-OCC as the optimal concurrency control protocol. It explores the limitations of GPU-2PL and GPU-OCC, and proposes future work on workload distribution algorithms. Additionally, it covers the benefits of proportional caching in reducing memory consumption for query throughput, and the concept of scaling analytical workloads with working set size. The lecture concludes with insights on SSB execution time, memory utilization, and the potential of lock-free GPU transactions.