A novel distributed data access scheme considering with link resources and metric in lambda grid networks

A link load balancing technique that uses a normalized link metric based on a new wavelength utilization and link metric is proposed. The new normalized link metric is a combination of a previous link metric and wavelength utilization. The shortest path algorithm selects the lowest normalized link metric, so that heavily loaded links are avoided if possible. Due to the development of WDM optical network technologies, a lambda grid system, which shares computing resources across a wide area optical network, has been proposed as a grid system. To reduce the influence of file damage and/or prevent particular storage servers from being over loaded, duplicate files are stored in several replica servers. Parallel downloading from these servers decreases downloading time and achieves load distribution. However, parallel downloading can create bottleneck links in the WDM network, since many connections may become used in common. In this paper, we propose a novel server selection method for parallel downloading in WDM networks. The server selection method creates link load balancing and suppresses bottleneck links. Computer simulations show that the proposed scheme can reduce bottleneck link number and the blocking probability by up to 90%.