An Empirical Study of Power Aware Load Balancing in Local Cloud Architectures

In this paper, resource load balancing is empirically investigated from the perspective of power consumption and resource availability. An experiment was conducted to investigate the impact of introducing a power aware load balancing algorithm on availability and operational cost of a local cloud as it scales to size. The experiment involved a comparison of a conventional round robin approach to load balancing with a proposed new power aware load balancing algorithm. The results have shown that there is significant cost savings while implementing the power aware algorithms. There also may be a trade-off in cost savings versus availability. In our experiment, a user may incur a waiting period as a compute node powers on. Our algorithm uses a best effort approach to reduce the latency on availability of resources.