Date Added: Jan 2010
With the constantly changing and regularly evolving applications that data centers have to deal with, the servers of these data centers are met with high workloads, which are often extremely demanding. This is primarily enhanced because of changing dynamics of online internet applications such as for e-retail, online banking, finance, news, social networking and communication. This paper seeks to address this issue and proposes a mix-aware technique that enables the provider to take care of dynamic workload in a data center while it is able to meet altercations in the volumes that flow in. This mix-aware technique is particularly useful when allocating server capacity in Internet data centers. The technique discussed in this paper uses the k-means clustering algorithm so that the data center can automatically figure out the ideal workload mix and a queuing model so that it is able to predict the server capacity for an imminent workload mix. Along with the technique, the paper implements a prototype provisioning system that can deploy the technique towards a productive end. The paper has also evaluated the efficiency of the mix-aware technique on a laboratory Linux data center that uses a TPC-W web benchmark. The experiment proved that this clustering technique is able to understand workload mix changes with Internet applications. This prototype also proves that the technique gets rid of SLA violations because of under-provisioning that happens in the case of dynamic web workloads.