학술논문

Virtual infrastructure planning: The GEYSERS approach
Document Type
Conference
Source
2012 Future Network & Mobile Summit (FutureNetw) Future Network & Mobile Summit (FutureNetw), 2012. :1-9 Jul, 2012
Subject
Computing and Processing
Communication, Networking and Broadcast Technologies
Fields, Waves and Electromagnetics
Signal Processing and Analysis
Planning
Optical fiber networks
Computer architecture
Topology
Resource management
Electronic mail
Virtual Infrastructure Planning
NDL
Evolutionary Game Theory
Information Modeling Framework
Language
Abstract
The new and emerging IT services require very high network capacities and specific IT resources that cannot be intrinsically delivered by the current Best Effort Internet. In response to this the European project GEYSERS (Generalised Architecture for Dynamic Infrastructure Services) is proposing a novel architecture that employs optical networking, capable of provisioning “Optical Network and IT resources” for end-to-end service delivery. GEYSERS adopts the Infrastructure as a Service framework and the Service-Oriented Networking paradigm and proposes an architecture that enables infrastructure operators to virtualize their infrastructures (optical network and IT resources) and offer them as a service based on the user/application requirements. This paper provides an overview of the GEYSERS approach regarding virtualization of infrastructures comprising optical network and IT resources. Special emphasis is given in the description of the Logical Composition Layer of the architecture that is responsible for both the creation and maintenance of virtual resources and the virtual infrastructures. An important function of the Logical Composition Layer is the virtual infrastructure planning process discussed in detail. An optimization scheme suitable to adaptively plan and re-plan virtual infrastructures employing evolutionary game theory is presented and compared to conventional centralized approaches. Our evolutionary game theory modelling results clearly indicate, that given sufficient time to learn the status of the underlying physical topology the virtual infrastructures planned have similar performance to those generated through traditional global optimization approaches such as integer linear programming.