Competitive analysis for service migration in VNets

  • Authors:

  • Marcin Bienkowski;Anja Feldmann;Dan Jurca;Wolfang Kellerer;Gregor Schaffrath;Stefan Schmid;Joerg Widmer

  • Affiliations:

  • Institute of Computer Science, University of Wroclaw, Wroclaw, Poland;T-Labs / TU Berlin, Berlin, Germany;Docomo Labs Europe, Munich, Germany;Docomo Labs Europe, Munich, Germany;T-Labs / TU Berlin, Berlin, Germany;T-Labs / TU Berlin, Berlin, Germany;Docomo Labs Europe, Munich, Germany

  • Venue:
  • Proceedings of the second ACM SIGCOMM workshop on Virtualized infrastructure systems and architectures
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

Network virtualization promises a high flexibility by decoupling services from the underlying substrate network and allowing the virtual network to adapt to the needs of the service, e.g., by migrating servers or/and parts of the network. We study a system (e.g., a gaming application) where network virtualization is used to support thin client applications for mobile devices to improve their QoS. To deal with the dynamics of both the mobile clients as well as the ability to migrate services closer to the client location we advocate, in this paper, the use of competitive analysis. After identifying the parameters that characterize the cost-benefit tradeoff for this kind of application we propose an online migration strategy. The strength of the strategy is that it is robust with regards to any arbitrary request access pattern. In particular, it is close to the optimal offline algorithm that knows the access pattern in advance. In this paper we present both an optimal offline algorithm based on dynamic programming techniques to find the best migration paths for a given request sequence, and a O(¼ log n)-competitive migration strategy MIG where ¼ is the ratio between maximal and minimal link capacity in the substrate network for a simplified model. This is almost optimal for small ¼, as we also show that there are networks where no online algorithm can achieve a ratio below ©(log n/log log n). In contrast, the optimal solution without migration can only achieve a competitive ratio that is linear in the network diameter. Our simulations indicate that the competitive ratio of MIG is robust to the network size, and that the ratio is small if the request dynamics are limited and the requests are correlated.