Start Date
11-8-2016
Description
In the domain of enterprise applications, operational costs can be reduced by consolidating orthogonal workloads with the objective of maximizing server utilization levels and minimizing the total amount of required capacity. This is closely related to the well-known bin packing problem which is NP-hard. Related problem formulations often consider varying historical workload traces, but include only one resource dimension, usually the CPU. This implicates a serious risk of overloading other resources that are not related to CPU demands, such as memory. Therefore, we formulate the multidimensional workload consolidation problem and develop eight algorithms to provide solutions. We evaluate their applicability using workload traces gathered from four data centers. A best-fit heuristic that uses a genetic algorithm provides best solution qualities with lowest variance and revealed up to 53.39 percent of unused capacity. In general, multidimensional workload consolidation problems eliminate less server capacity, but effectively reduce the risk of resource overloads.
Recommended Citation
Müller, Hendrik and Bosse, Sascha, "Multidimensional Workload Consolidation for Enterprise Application Service Providers" (2016). AMCIS 2016 Proceedings. 7.
https://aisel.aisnet.org/amcis2016/EntSys/Presentations/7
Multidimensional Workload Consolidation for Enterprise Application Service Providers
In the domain of enterprise applications, operational costs can be reduced by consolidating orthogonal workloads with the objective of maximizing server utilization levels and minimizing the total amount of required capacity. This is closely related to the well-known bin packing problem which is NP-hard. Related problem formulations often consider varying historical workload traces, but include only one resource dimension, usually the CPU. This implicates a serious risk of overloading other resources that are not related to CPU demands, such as memory. Therefore, we formulate the multidimensional workload consolidation problem and develop eight algorithms to provide solutions. We evaluate their applicability using workload traces gathered from four data centers. A best-fit heuristic that uses a genetic algorithm provides best solution qualities with lowest variance and revealed up to 53.39 percent of unused capacity. In general, multidimensional workload consolidation problems eliminate less server capacity, but effectively reduce the risk of resource overloads.