As a timely contribution to energy efficiency challenges across data centers and clouds, the authors propose a virtual machine (VM) consolidation approach to limit the number of physical machines provisioned and active. By dealing with the problem of optimized component provision, the authors have a view to optimize the number involved in a computation as a modern application of dataflow graphing in VM provision.

The approach correlates the computation, storage, and network resource demands of one VM with the subsequent resource demands of others. Workload intensity is additionally correlated with VM resource demand. Together, it is recognized that resource demands are dependent on one another and, as different resources may be bottlenecks on performance, both correlations and resource types are considered.

Demands are modeled as random variables, with randomness depending on the applications running. This approach overcomes the memory requirements of a correlation coefficiency matrix, with demands modeled as linear functions of random variables. Assumptions on the probability distributions of resources consumed, however, are not substantiated by real data. In their solution, the least correlated VMs are clustered so that overall cluster use is minimized.

Hwang and Pedram work to show that a multi-capacity stochastic bin packing (MCSBP) problem is nondeterministic polynomial-time (NP)-hard. They consider this to be a variation of the bin packing problem, which is used as the NP-complete problem, and express that the bin packing problem is reducible to MCSBP.