Experiments with Smart Workload Allocation to Cloud Servers

Abstract

We present experiments that compare three on-line real time techniques for task allocation to different cloud servers: an adaptive random neural network (RNN) based on reinforcement algorithm, an algorithm based on "sensible routing'', one which uses a simple analytical model to select the server is estimated to give the best response as a function of workload, and round-robin task allocation. Measurements indicate that the RNN based algorithm can make accurate decisions when it exploits frequent measurement updates.