Processing OLTP Workloads on Hybrid CPU/GPU Systems