Section: New Results
Monitoring of MESH networks
Participants : Olivier Festor, Emmanuel Nataf [ contact ] , Cristian Popi.
Related to monitoring aspects in wireless mesh networks, we have proposed, evaluated and implemented a prototype of WiMFlow, a distributed and self-organized flow monitoring framework for wireless mesh networks. Since flows entering/exiting a wireless mesh network, usually take multiple hops inside the backbone of the mesh network to reach the destination, a naive approach could have all backbone nodes (mesh routers) monitor the flows and export the collected flow information to a collector. This results in redundant flow information being sent by multiple mesh nodes, which entails a high export network overhead.
In our work we searched for mechanisms to distribute the flow monitoring charge accross the network, in such a way that minimizes the number of times a flow is monitored, while adapting the monitoring service availability to the dynamicity of the network's backbone due to link instability or node shut-down inherent in wireless mesh networks.
All routers behave as possible probes. In order for the probes to make decisions on which one monitors a flow, a global vision of the routing entries of all the nodes in the backbone is required on every node. This allows a probe that sees a flow passing through its interfaces to trace the flow's path. A prerequisite for monitoring a flow is that a probe P that sees a flow F on one of its interfaces has to know the flow's entry and exit points in the backbone, as well as the next hop towards the exit point for each node on the path of the flow.
In accordance with this, we proposed a multi-layered functional architecture of the monitoring system. The routing plane builds up the routing table of the mesh nodes (with the help of a pro-active routing protocol). It then provides the routing table entries of all nodes to the monitoring overlay, which uses this information to organize the nodes into monitoring or non-monitoring probes.
Two components come into the decision making process when organizing the nodes for monitoring: the routing information received from the routing plane to locally build the path of a flow on a node, and the metrics that allow to differentiate between nodes located on the path of the flow. These metrics are distance (in number of hops) of the node from the collector (to which the node is configured to send flow records), connectivity degree and up link quality. Nodes with better distance, higher connectivity or up link quality are the ones elected to monitor the flow.
In order to reduce the number of control messages we use the concept of Multi Point Relay (MPR) employed by the OLSR routing protocol to flood topology control messages. The MPR Set selection scheme is that of OLSR. Hello messages are used to convey neighbourhood information. For flooding the network with routing entries, routing control message (RC) are sent by every node and broadcast via the MPRs, containing the entire routing table of the sender. In a second phase we have proposed a mechanism to adapt the emission times of the Hello and RC messages to the dynamicity of the topology, with the goal of keeping the cost of the monitoring overlay low.
A modular implementation prototype of WiMFlow has been implemented based on nprobe for flow information packaging into v5 and v9 export formats, which was tested on a small-scale wireless mesh network set-up in the premises of the team's offices.
In the area of configuration management, we have also worked on the semaless integration of Netconf-based XML oriented management with the new data-modeling language under standardization within the IETF: YANG. We were the first to offer a full Yang-based operational manager demonstrating the usefulness of this approach.
