Supporting bulk data transfers of high-end applications with guaranteed completion time

Abstract

In high-end grid networks, distributed resources (scientific instruments, CPUs, storages, etc.) are interconnected to support computing-intensive and data-intensive applications, which often require reliable and efficient transfer of gigabits or even terabits of data among endpoints. To facilitate the efficient scheduling of endpoint resources, networks should provide a service which guarantees that a specified amount of data is transferred within a strict time window. Existing "FixRate-FixTime" resource reservation architectures such as IntServ/RSVP reserve for each flow a constant amount of resource from a specified starting time. Instead, a bulk data transfer can start at any time after its arrival, use any and even time variant bandwidth value, as long as it is completed before its deadline. This paper proposes a flexible "Multi-Interval" resource reservation approach, which divides the active window of a transfer into multiple sub-intervals, and reserves constant amount of resource in each of them. We identify a possible evolution path towards the proposed architecture, and use simulations to demonstrate the potential performance gain from increasing flexibility. © 2007 IEEE.