Local Adaptive Resource Scheduling for Internet-Based Computation on SMT Platform

This work presents an adaptive resource scheduling policy (ARS) for local workloads of internet-based computation on SMT platform, which can dynamically adjust resource assignment among threads according to changing workloads behavior and real-time performance by taking in-flight performance as feedback to drive resource assignment decision. The adaptive mechanism and similarity control of assignment candidates function together to impulse persistent resource scheduling in changing program environment. Simulation results show that ARS achieves much improvement both in throughput and fairness metric. Meanings and settings of important ARS parameters are also discussed.