Rerouting in advance reservation networks

Abstract

The advance reservation of network connections is an area of growing interest and a range of service models and algorithms have been proposed to achieve various scheduling objectives, i.e., including optimization-based strategies and heuristic schemes. Now given the time-shifted nature of future requests, rerouting strategies have also been considered to improve resource allocation and carried loads. However, most existing rerouting schemes have focused on minimizing connection disruptions and have not considered further link load information. Along these lines, this paper develops novel rerouting strategies to improve connection scheduling in advance reservation networks. Specifically, a dynamic optimization formulation is presented to handle “on-line” arrivals along with a new heuristic load-balancing strategy. The performance of these proposed solutions is then evaluated for a wide range of network topologies and also compared against some existing rerouting schemes.

Introduction

Advance reservation (AR) of user connections is becoming a key requirement in next-generation networks. As compared to regular immediate reservation (IR) provisioning, AR provides operators with a powerful tool to improve the usage of their deployed infrastructures and boost revenue potential. Now many users in the “e-science” field already require “pre-scheduled” network capacities to support applications such as large-scale data transfers, workflow process management, and distributed grid/cluster computing [1], [2]. Here, given the massive increase in scientific data volumes, AR is particularly imperative as even the most scalable backbones may not be able to support all requests in an on-demand manner. The topic of AR scheduling has received much focus in recent years, with a range of algorithms studied for bandwidth-provisioning and optical wavelength routing networks [3], [4], [5], [6], [7], [8], [9], [10], [11]. Some of these studies have also considered broader protection/survivability issues [9] as well as flexible service models, e.g., such as variable start/stop times, variable capacity requirements, etc. [10], [11]. Furthermore, researchers have also applied rerouting concepts (originally developed for IR networks [12], [13], [14], [15], [16]) to achieve improved resource distribution for AR demands [17], [18], [19], [20]. These methods are particularly attractive since non-active future reservations can be moved in a non-disruptive manner. However, for the most part, existing AR rerouting schemes have used graph-based heuristic strategies and have focused on minimizing the disruption of existing reservations (caused by rerouting). Although this is a desirable objective, the further use of resource optimization techniques can be very beneficial here. For example, link load information can potentially be very beneficial for achieving improved network load distributions and thereby improving the number of AR requests scheduled in the network. Along these lines, this paper presents a more comprehensive look at AR rerouting, using formalized ILP methods to improve resource distribution. The key aim here is to incorporate link load information in order to increase network resource efficiencies and increase provisioning success rates. The paper is organized as follows. First, Section (2) presents a background survey of existing AR rerouting schemes. Section 3 then presents an ILP formulation, including a novel “dynamic” rerouting scheme to handle sequential “on-line” requests. Meanwhile a counterpart load-balancing rerouting heuristic is detailed in Section 4 and its run-time complexity analyzed. Finally, detailed performance evaluation results are presented in Section 5 for a range of network topologies and scenarios and comparisons made against some existing rerouting schemes. Overall conclusions and directions for future work are then presented in Section 6.