Towards Energy Proportionality for Large-Scale Latency-Critical Workloads

   Abstract