Scalable Service Migration: The Tree Topology Case (bibtex)
by Konstantinos Oikonomou, Ioannis Stavrakakis
Abstract:
Scalable service placement is a challenging problem in dynamic environments such as ad hoc or autonomic networks. Existing approaches typically consider static and reduced size networking environments and try to determine the optimal service position (the node at which some cost is minimized) by solving the 1- median problem. Since such approaches are complex and are based on global information knowledge, they are non- scalable and cannot cope with dynamic environments. A more reasonable approach to service placement for large, ad hoc and autonomic environments would be through service migration. That is, instead of solving continuously a large optimization problem requiring global information, consider policies for moving the service position (one hop/node at a time) based on local information, towards more effective positions. Developing service migration policies with good properties is a major challenge, since such policies may be sub-optimal (that is, they never converge to the optimal position), follow a non-monotonically cost decreasing path to the optimal position, etc. In this paper, the aforementioned issues are discussed and a simple service migration policy is proposed for undirected tree topologies. For this case it is shown analytically that the information available at the current service node only is sufficient for determining the direction towards nodes with monotonically decreasing service provision costs. Consequently, the proposed policy moves the service continuously towards the optimal position in every step and reaches the optimal position through a shortest path migration trajectory. As the optimal position may change in a dynamic environment, the proposed policy adapts the service migration path continuously towards the currently optimal position. Although some of the main results are also applicable to general network topologies, future work will focus on the general network topology by borrowing ideas from the current work.
Reference:
Konstantinos Oikonomou, Ioannis Stavrakakis, "Scalable Service Migration: The Tree Topology Case", In 5th Annual Mediterranean Ad Hoc Networking Workshop, 2006. ([pdf])
Bibtex Entry:
@inproceedings{oikonomou2006scalable,
	Abstract = {Scalable service placement is a challenging problem in dynamic environments such as ad hoc or autonomic networks. Existing approaches typically consider static and reduced size networking environments and try to determine the optimal service position (the node at which some cost is minimized) by solving the 1- median problem. Since such approaches are complex and are based on global information knowledge, they are non- scalable and cannot cope with dynamic environments. A more reasonable approach to service placement for large, ad hoc and autonomic environments would be through service migration. That is, instead of solving continuously a large optimization problem requiring global information, consider policies for moving the service position (one hop/node at a time) based on local information, towards more effective positions. Developing service migration policies with good properties is a major challenge, since such policies may be sub-optimal (that is, they never converge to the optimal position), follow a non-monotonically cost decreasing path to the optimal position, etc.
	In this paper, the aforementioned issues are discussed and a simple service migration policy is proposed for undirected tree topologies. For this case it is shown analytically that the information available at the current service node only is sufficient for determining the direction towards nodes with monotonically decreasing service provision costs. Consequently, the proposed policy moves the service continuously towards the optimal position in every step and reaches the optimal position through a shortest path migration trajectory. As the optimal position may change in a dynamic environment, the proposed policy adapts the service migration path continuously towards the currently optimal position. Although some of the main results are also applicable to general network topologies, future work will focus on the general network topology by borrowing ideas from the current work.},
	Author = {Oikonomou, Konstantinos and Stavrakakis, Ioannis},
	Booktitle = {5th Annual Mediterranean Ad Hoc Networking Workshop},
	Comment = {<a href="./MyPublications/oikonomou2006scalable.pdf">[pdf]</a>},
	Keywords = {own, refereed, ana},
	Title = {{{Scalable Service Migration: The Tree Topology Case}}},
	Venue = {Lipari, Italy},
	Year = {2006}}
Powered by bibtexbrowser