Unity metric based highly adaptive scheduler for smartphones (UMBASS)

Since the introduction of Pervasive Computing, there has been a need for a scheme that partitions computation among local CPU and server for optimal performance without compromising energy consumption. In order to adapt to rapidly changing network and computational resource, partitioning schemes evolved from static to dynamic. A recent approach proposed to move scheduler from server to client resulting in better adaptability hence, improved performance. This work presents UMBASS, an improved mobile device installed scheduler. UMBASS introduces unity (single) metric based scheduling supporting configurable priority for the metric. Design of UMBASS requires minimal computations due to single metric based RPC level scheduling thus making it highly adaptive. The evaluation methodology used in this work is general and can serve as a framework for evaluation of partitioning schemes for smartphone applications. Based on this framework, trace-based simulation results show that UMBASS achieves average speedup gain of 24% over previous best client-server partitioning scheme and 19% over previous best mobile installed dynamic partitioning scheme. Average energy savings are on par with these schemes.