Data Centre Optimisation and Bin Packing: An Overview
In this talk we present different optimisation problems related to the operation of data centres and virtual machine assignment, and study their relation to bin packing. The use of virtual machines for packaging user services has allowed a large improvement in running data centres more efficiently. Multiple virtual machines can be packed on the same physical server, allowing efficient use of computing resources, while minimising any interference between the applications. By moving virtual machines between servers we can reduce the overall energy cost as the work load changes during the day, switching off machines or even complete parts of a data centre in slack times. Moving services between data centres allows to exploit temperature differences between day and night, and between geographical zones, and also gives an opportunity to minimise electricity cost due to time variable prices. In some situations, we can control which tasks are accepted to be run (demand acceptance), or schedule runs at times that best utilise available resources. Underlying the models for many of these problem variants is the bin packing global constraint and its variants, or versions of the cumulative constraint.
Researcher at Insight Centre for Data Analytics
Wednesday, 16 December, 2015 (All day)