The Community for Technology Leaders
2009 Sixth International Conference on Broadband Communications, Networks, and Systems. Broadnets 2009 (2009)
Sept. 14, 2009 to Sept. 16, 2009
ISSN: 2150-4903
ISBN: 978-963-9799-49-3
pp: 1-8
Ashwin Gumaste , Indian Institute of Technology, IIT Bombay, India 400 076
Shakesh Jain , Akamai Inc. Santa Clara, CA. USA
Arun K. Somani , Iowa State University, Ames, USA
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.
bandwidth allocation, scheduling, wavelength division multiplexing

A. Gumaste, S. Jain and A. K. Somani, "An efficient superscheduler architecture and job migration algorithm for computational grids over light-trail WDM networks: Invited paper," 2009 Sixth International Conference on Broadband Communications, Networks, and Systems. Broadnets 2009(BROADNETS), Madrid, 2009, pp. 1-8.
95 ms
(Ver 3.3 (11022016))