Minimising the size of voids in optical packet switching using a novel packet scheduling algorithm

Abstract

In this paper, we propose a new packet scheduling algorithm to minimise the size of voids in optical packet switching. The effects of excess load, output utilisation, and packet loss probability are closely studied. Other contributing factors to the packet loss probability, which include the buffer depth and the numbers of wavelength channels, are also investigated. The proposed algorithm is of importance to next generation networks where broadband capabilities with end-to-end quality of service over all-IP optical networks is envisaged. IntroductionWhen optical packet switching (OPS) was first proposed in the research community, it was hoped that it would become a remedy to the shortcoming of electronic packet switching, handling large bandwidths with minimal intelligence [1,2]. The early proposals by RACE ATMOS, ACTS KEOPS, and others that followed, were based on the assumption that ATM would be the dominant technology at the time optical packet switches were deployed. However, as time passes, ATM has not been as popular as had been anticipated. Instead, the future of ATM is being challenged by the introduction of MPLS in the WAN and Ethernet in the LAN. Consequently, optical packet switching, with the capability to handle variable-length packets, is being introduced as an alternative to ATM technology. Processing variable length packets optically requires intelligence in the switching process in order to reduce packet loss probability [3]. In order to increase the network utilisation, it is necessary to decrease the size and frequency of voids by increasing the buffer granularity. To further decrease the void sizes, ACK packets are scheduled to fill in additional gaps.In this paper we consider the reasons for void formation and propose a buffering mechanism with a new ACK scheduling algorithm. Simulations of the new algorithm are presented and the results discussed. Void formingIn IP networks, having variable packet lengths (rather than fixed length packets) results in packet synchronisation processes at input ports becoming much more complicated and difficult. One of the problems with buffering variable-length IP packets is the setting of optical delay lines to match the sizes of the packets. Since the sizes of IP packets vary in a wide range from 40 bytes to around 1500 bytes, it is practically impossible to set up the delay lines to have the same granularity (the step increase of delay lines which are assumed to be uniform) and at the same time have a reasonable buffer depth (the number of delay lines). Failing to match the sizes of the IP packets to the delay lines will create unnecessary spaces at the output link, referred to as voids, and causes high excess utilisation (defined as a ratio of the generated voids in bytes over the total transmitted bytes [4]). The voids and excess utilisation, is a result of the delay line mismatch; however, such phenomena can also be caused by the difference in arrival times between packets from different sources, which is referred to as arrival time mismatch. Figure 1 il...