Performance-driven calibration of ERGO - Springer

摘要

Load-balancing research in unstructured peer-to-peer networks has been propelled by the need to improve performance and efficiency whilst retaining the desirable properties of these networks. This work builds further on an already researched efficient load-balancing scheme called Enhanced-Reconfigurable-Gnutella-Overlay (ERGO), utilising a concept of virtual-server-overlay on Gnutella networks for the purpose of load-balancing. The server-node ratio is an important aspect of these networks as they enhance or impair the functionality of the network and thus the main focus of our research. Firstly, the influence of various parameters such as network load, Node Capacity, Server Capacity and the corresponding effects of server-node ratio on metrics such as Success of Queries, Balanced Nodes and the Cost of achieving the balance were studied. Secondly, an optimum server-node ratio was ascertained, optimum implying the maximum Balance at minimum Cost and high Query Success. Findings using a purpose-built simulator, supported an increased ratio of servers with load, and conversely a decreased ratio for discovering the ERGO server. Experiments showed the optimum server-node ratio to be at 5%.