Risk Aware Resource Allocation for Clouds∗

Cloud computing offers on-demand access to large-scale computing resources in a pay-as-you go manner. Market-based resource allocation mechanisms are gaining popularity among commercial cloud providers to deal with dynamically fluctuating resource demands. For example, the recently introduced Amazon EC2 spot instances allow users to bid for computing resources and thus control the cost vs. reliability trade-offs of their workloads. Although this promises significant cost reduction, it comes at an additional risk of price fluctuation. This will get worse as cloud computing gradually moves towards a free market system. We propose a novel approach that utilizes financial option theory to simultaneously mitigate risk and minimize cost for cloud users. We formulate the cloud user optimization problem and mathematically characterize the cost of using European style options for clouds. We also propose a novel on-line policy using American options that outperforms base-line spot policies in terms of price variance reduction against high risk factors. We present trace-driven simulation experiments to support our results.