Network fault monitoring in Grid

Abstract

Grid resources having heterogeneous architecture being geographically distributed and interconnected via unreliable network media are at the risk of failure which proves the need for an efficient fault monitoring framework. The traditional network fault monitoring systems based on the centralized client/server architecture have limited efficiency and scalability, as the complexity of the network increases, but the mobile agents with specific functions can be dispatched to network nodes and accomplish the assigned tasks. The mobile agent based model provides efficiency and flexibility in network fault monitoring, since dispatched agents avoid unnecessary traffic overheads due to frequent data transmissions between the compute nodes and the head node in a cluster and this model can be used in clusters of any size. The proposed system involves monitoring network related faults in a Grid environment. The network related faults covered in this system are link failure, network traffic overloads and resulting packet losses. Both the link failure and the packet loss due to congestions in the network, prevents the corresponding application from proceeding further which results in delay in job completion. Overload in network traffic which occurs due to congestions caused by packet flow exceeding the maximum network throughput will further result in packet losses and delays in network flow which increase the job completion time. Detecting these network failures can help in better utilization of the resources and timely notification to the user in a Grid environment.