Energy-efficient simulation and performance evaluation of large-scale data centers

This paper presents a methodology and a tool for modeling and simulating job assignment and migrations in large scale cloud infrastructures consisting of hundreds of thousands of processing, storage and networking nodes. Each cloud node, whether a server, or a disk array or a network element can be modeled according to a generalized single node queuing model, with appropriate parameterization and multiple job class definitions. The queuing model is solved for each node using an approximate mean value analysis technique (AMVA). The solver computes resource utilizations, response times, throughputs and delays and identifies bottlenecks. It is very fast, parametric and scalable to suit the analysis of large scale cloud infrastructures and data centers or server farms. An interactive and batch model solver and simulator have been developed to simulate job assignment and consolidation for energy efficiency, by solving the proposed model for up to 500.000 cloud nodes and several millions of jobs in a few minutes. SLAs and virtual memory restrictions are optionally considered, too. The scalability and speed of this cloud modeling technique and model solver make it a unique tool for studying problems and algorithms related to job migrations for very large cloud infrastructures. A sample set of preliminary experimental results are presented to validate the behavior of the model and the tool.