Edinburgh Napier University

  Distributed Knowledge Management Systems (DKMS) often depends on the Semantic Web Peer-to-Peer (SW-P2P) model. The reason for this is based on its support for autonomy of knowledge node, ease of accessibility and scalability. The susceptibility to failure experienced during knowledge retrieval has been a concern for the SW-P2P. This paper presents a fault tolerance system in order to resolve the problem of the DKM. The architecture of this design consists of five components namely; Replication Manager (RM), Fault Detector (FD), Fault Notifier (FN), Recovery Mechanism (RMe) and Global Control Monitor (GCM). This design adopted dynamic replication strategy and group constitution procedure to guarantee knowledge availability on knowledge nodes. The dynamic replication strategy was used to create and delete replicas based on the changes in the DKMS environment. The group constitution procedure suggested the efficiency of fault recovery process in terms of the best available replica among knowledge service group. The fault tolerance system execution cycle was performed on a set of Virtual Machines (VM) using the VMware Workstation version 7.0.1, while Java programming language was used to implement the group and ungroup replicas. Sample data of varying magnitude in ranges of 225Kilobytes to 512Kilobytes and 450Kilobytes to 512Megabytes were tested at different time intervals on both the grouped and ungrouped replicas at a threshold between 0.85 and 0.9 of knowledge retrieval. The results showed a reduction in the average response time of the grouped replicas which was measured to be 34 milliseconds and 68.2 milliseconds against un-grouped replica that was estimated as 53 milliseconds and 107.2 milliseconds respectively. The effect of this reduction in response time was that the grouped replica was faster than the approach of un-group replica. In addition, the group replica occupied less memory space because it does not need to store replicas on the active knowledge peer when recovering from failure. This result showed that the system guarantees the fault tolerance of each knowledge node in a DKMS.