S-CORE: Scalable Communication Cost Reduction in Data Center Environments

Tso, F. P., Oikonomou, K., Kavvadia, E., Hamilton, G. and Pezaros, D. P. (2013) S-CORE: Scalable Communication Cost Reduction in Data Center Environments. Technical Report. School of Computing Science, University of Glasgow.

Abstract

Maximizing return-on-investment through efficient resource usage of Data Center (DC) infrastructures is of paramount importance for Cloud service providers. Although a number of networking architectures and routing/flow scheduling protocols have been developed to offer full bisection bandwidth over such topologies, deployment costs and machine virtual- ization can impact the aggregate network load and result in episodes of congestion, especially at the core. In this paper, we propose S- CORE, a scalable Virtual Machine (VM) migration strategy to dynamically allocate VMs to servers so that the overall communication footprint of the active traffic flows is minimized. We first formulate the aggregate VM communication as an optimization problem whose solution minimizes a communication cost function. This optimal yet centralized approach is of high complexity and requires global information, constituting it not scalable for inherently dynamic DC environments. We then define a distributed migration policy based on local information that adapts to dynamic traffic changes and achieves significant communication cost reduction. We evaluate S-CORE over diverse aggregate load, DC topologies, and coordination policies, and we show that it can achieve up to 90% communication cost reduction, while its locality properties can be exploited for other purposes such as energy footprint reduction.

• Item Type: Research Reports or Papers (Technical Report)
• Additional Information: <br>DCS Tech Report Number: TR-2013-338</br> <br>Paper ID: 9397</br> <br>Appeared in: SoCS Technical Report Series</br>
• Status: Published
• Glasgow Author(s) Enlighten ID: Tso, Dr Fung Po and Pezaros, Professor Dimitrios and Hamilton, Mr Gregg
• Authors: Tso, F. P., Oikonomou, K., Kavvadia, E., Hamilton, G., and Pezaros, D. P.
• College/School: College of Science and Engineering > School of Computing Science
• Publisher: School of Computing Science, University of Glasgow