COMMUNE (COgnitive network ManageMent under UNcErtainty) is a Celtic project funded from November 2011 to May 2014. COMMUNE aims to build an innovative solution for cognitive network management under uncertainty, and will seek to mitigate the practical effects of uncertainty by exploring the latest advances in knowledge based reasoning and other relevant cognitive methods.
COMMUNE concepts can be grouped along the well-known MAPE (Monitor-Analyze-Plan-Execute) loop. Figure 1 shows the COMMUNE vision summarizing the different areas of the COMMUNE approach. It is important to emphasize that COMMUNE deals with uncertain information at various levels in the network hierarchy (individual network elements, Network Management Systems) and at various locations in the network, thus enabling distributed, centralized and hybrid Cognitive Network Management solutions.
COMMUNE aims at building an innovative solution for the management of Future Networks under uncertainty, i.e., COMMUNE will apply mechanisms to reduce uncertainty associated with critical management situations. In COMMUNE Project, AIT will focus on QoE management for optimising scalable video streaming over a P2P network. The goal is to optimise QoE for users based on the super-peer deployment approach, which aims to maximise QoE aggregated across all peers whilst minimizing the volume of content each peer receives from the super-peers.
The Figure 2 describes an overview of the system with cloud backed super-peers deployed in different geographical regions supporting content distribution in the P2P network. Super-peers deployed in this manner facilitate time and region diversities for content in a globalised way. For instance the demand for video content for a particular stream may vary from region to region as peak usage time’s change. By deploying cloud instances as super-peers we take advantage of the globally distributed nature of cloud computing facilitating improved streaming performance. To facilitate optimal super-peer deployment in P2P streaming sessions, we employ the Q-learning algorithm, which is a model free method which attempts to approximate a control policy directly through environmental actions, to aim at tackling the planning problem under uncertainty.
Brian Lee, Chunrong Zhang, Yuansong Qiao
Athlone Institute of Technology, Ireland
Daysha Consulting, Ireland
Elisa Corporation, Finland
Magister Solutions, Finland
Nokia Siemens Networks Oy, Finland
Oy L M Ericsson Ab Finland, Finland
Telekomunikacja Polska S.A., Poland
Telenium, Tecnologia y Servicios S.L., Spain
Telnet Redes Inteligentes, Spain
VTT Technical Research Centre of Finland, Finland