An efficient superscheduler architecture and job migration algorithm for computational grids over light-trail WDM networks: Invited paper

The merging, management and utilization of pervasive, idle computing devices using a dynamic communication infrastructure leads to the concept of computational grids. A computational grid enables enterprises to efficiently use distributed computing entities in a cost-effective setup for emerging compute-intensive applications. From a network perspective, the key requirement for a computational grid is the associated ability to provide dynamic bandwidth (on-demand). An optical networking solution based on the light-trail concept readily adapts to the requirement of dynamic provisioning of bandwidth, in addition to being a low-cost solution and providing for optical multicasting - the key to distribution of jobs amongst multiple nodes without the overhead of processing. We investigate a protocol based on a double auction mechanism for converting a light-trail based network into a computational grid using superschedulers at each node. It allows job migration to idle processors. We show how the auction scheme works for bandwidth assignment and in the technique of computing bids, as well as alterations in the virtual topology of the light-trail network. Results of simulation study show the methods' effectiveness.