Quality of service differentiation, teletraffic analysis and network layer packet redundancy in optical packet switched networks

• Main Author: Øverby, Harald
• Format: Doctoral Thesis
• Language: English
• Published: Norges teknisk-naturvitenskapelige universitet, Fakultet for informasjonsteknologi, matematikk og elektroteknikk 2005
• Subjects: Telekommunikation; Telecommunication
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:no:ntnu:diva-563; http://nbn-resolving.de/urn:isbn:82-471-7126-0

Optical Packet Switching (OPS) has emerged as a promising candidate for the next-generation Wavelength Division Multiplexed (WDM) based alloptical network. By enabling packet switching in the optical domain, OPS networks can provide cost-efficient and transparent transport services to higher layers. However, a commercial deployment of OPS requires not only a maturation of several key enabling technologies, but also a thorough investigation of a number of networking challenges related to OPS, since OPS networks are fundamentally different from today’s store-and-forward networks. This thesis addresses the latter issue by considering the following three OPS networking issues: · Quality of Service (QoS) differentiation at the WDM layer, with focus on packet loss rate (PLR) and delay-jitter differentiation. · Teletraffic analysis of OPS networks. · How to combat packet loss in OPS networks by using network layer packet redundancy. First, a crucial issue in OPS networks is packet loss at the network layer due to contention. Contention occurs when a packet is destined for a wavelength currently occupied by another packet. Several approaches to combat such packet loss have been proposed in recent literature, e.g. by utilizing wavelength conversion, buffering, deflection routing or traffic shaping. This thesis considers a novel approach to combat packet loss in OPS: The proposed Network Layer Packet Redundancy Scheme (NLPRS) allows redundancy packets to be injected into the OPS network, thus enabling reconstruction of lost data packets at the OPS egress node. Results show that the NLPRS is able to reduce the end-to-end data PLR several orders of magnitude in an asynchronous OPS ring network with and without wavelength conversion. Another crucial issue in OPS networks is QoS differentiation at the WDM layer. Due to the lack of optical random access memory, existing QoS differentiation schemes suitable for today’s WDM point-to-point architecture are not feasible to use in OPS networks. Hence, new schemes that utilize the WDM layer to provide QoS differentiation are needed. A preemption based QoS differentiation scheme, the Preemptive Drop Policy (PDP), has been proposed for asynchronous bufferless OPS. With the PDP, high priority arrivals are allowed to preempt and take over a busy wavelength currently occupied by a low priority packet in the case of contention. This results in a lower PLR for high priority traffic compared to low priority traffic. The PDP has been extended into the Adaptive PDP (APDP), which provides absolute guarantees to the PLR for high priority ivtraffic in OPS by using a measurement based preemption probability parameter adjustment. An access-restriction based QoS differentiation scheme, the Wavelength Allocation algorithm (WA), has been studied. In the WA, which provides QoS differentiation in asynchronous bufferless OPS networks with full range output wavelength converters, a certain number of wavelengths at an output fibre are exclusively reserved for high priority traffic. When QoS differentiation (with respect to the PLR) is introduced in asynchronous OPS, it has been shown that the average throughput decreases, often referred to as the throughput penalty of introducing QoS differentiation. The main cause for this throughput penalty is because network resources must be used in a non-optimal manner when employing QoS differentiation schemes that utilize the WDM layer to isolate the service classes. However, as shown in this thesis, the throughput penalty is only found in asynchronous OPS. For slotted OPS, the average throughput stays the same after the introduction of QoS differentiation. An evaluation framework suitable for quantifying the throughput penalty when introducing QoS differentiation has been proposed. Using this framework, three fundamental different QoS differentiation schemes for asynchronous OPS, including the PDP and the WA, have been evaluated. It has been shown that preemptive techniques result in the lowest throughput penalty, followed by access-restriction and dropping based techniques. This is because, when using preemption, packets are dropped only when the output port is congested. With access-restriction, packets are dropped when the output port is highly strained, and with statistically packet dropping, packets are dropped independently of the state of the output port. A QoS differentiation scheme for slotted OPS has been proposed and evaluated. The scheme isolates the service classes by ensuring that a certain number of high priority packets can be transmitted at an output port in a time-slot in the case of contention. Using the proposed scheme does not result in a reduced throughput when the service classes are isolated. QoS differentiation schemes for asynchronous OPS with a share pernode (SPN) contention resolution pool architecture consisting of Tunable Wavelength Converters (TWCs) and Fibre Delay Lines (FDLs) have been proposed. In particular, it has been shown that the PLR and delay-jitter may be independently differentiated in this switch architecture. Analytical models of some of the proposed QoS differentiation schemes have been derived, providing explicit results of the PLR. In addition, an analytical framework regarding packet arrivals to an output port in an optical packet switch has been derived for both asynchronous and slotted OPS. This framework is particularly useful for studying the effects of nonuniform traffic. Furthermore, it has been shown that both the Erlang and Engset traffic models are suitable to model packet arrivals to an output port in an asynchronous optical packet switch. Regarding the Engset traffic model, it has been shown how the blocking probability can be evaluated vusing either the Engset lost calls cleared (LCC) traffic model or the Engset overflow (OFL) traffic model. For all Engset based traffic models, the time-, call- and traffic congestion have been derived. A numerical evaluation of the presented traffic models reveals that there is a small, but non-negligible, deviation between the observed blocking probabilities, which depends on the number of input/output fibres and the system load.