Research On The Strategy Of Quality-of-Service In Optical Burst Switching Networks

At present, optical fiber communication networks have become the backbone of transport networks. With the development and maturity of the wavelength division multiplexing (WDM) technology that can provide vast bandwidth for information transmission, a kind of all-optical switching network is expected to realize in near future. In order to satisfy the increasing needs of multimedia traffics explosively in Internet network, various optical communication schemes are put forward and studied in recent years.Up to now, three kinds of optical switching schemes based upon WDM, i.e., optical circuit switching (OCS), optical packet switching (OPS) and optical burst switching (OBS), are proposed. In contrast with OCS and OPS, OBS is a promising all-optical data transport technique to effectively use the bandwidth. OBS combines the merits of OCS and OPS while avoiding their defects. It is easy to be realized and can support bursty traffics such as IP packets well.The main problem of OBS networks as a base of the next generation of optical Internet is whether it can provide differentiated services for different services according to their quality of service (QoS) requirements. That is, OBS must provide better QoS guarantee. Two issues about QoS should be paid attention to OBS networks. First, how to solve contentions between data bursts (DB) and decrease DB loss probability. Second, how to reduce DB end-to-end transport delay, including DB assembling time at edge nodes. Obviously, it is more difficult to deal with QoS issue in OBS networks than in IP networks. So it is an attractive subject to study QoS performance of OBS networks.In the thesis the QoS issues of OBS networks are investigated, a novel protocol reference model is suggested and three new methods for improving the performance of OBS network are proposed. By some theoretical analysis and computer simulations, it is shown that the proposed methods can improve QoS performances evidently.1. General purpose OBS protocol reference modelThe proposed OBS network architecture consists of data plane, control plane and management plane. The data plane is divided into four layers, which are called high layer (IP layer), burst adaptation layer (BAL), OBS layer and physical layer (WDM layer) respectively. BAL is responsible for traffics converging and disconverging, DB assembling and disassembling. OBS layer provides DB accessing and switching functions. Core nodes need accomplish the functions of WDM layer and OBS layer only, whereas edge nodes must fulfill all functions of the four layers. The control plane is divided into signaling high layer, signaling adaptation layer (SAL), OBS layer and WDM layer. The PDU of signaling high layer is signaling message, the PDU of SAL is BCP packet, and the PDU of OBS layer is BCP frame. In order to satisfy requirements of different signaling protocols signaling message and the corresponding BCP should have several types. In this way, the expansibility of OBS networks is enhanced remarkably.2. DB Delayed Scheduling AlgorithmUsually the contention between DBs in OBS networks can be solved by fiber delay lines (FDL) buffer equipped at core nodes to delay the latter of contending DBs, which is called equality scheduling algorithm. We propose a novel scheduling algorithm called DB delayed scheduling algorithm. The main idea is that when there is a contention happening at a core node, the low-priority DB is buffered with FDL in order to forward the high-priority DB as early as possible, and the reserved wavelength channel for the low-priority DB will be re-assigned to the contending high-priority DB. Then a new BCP, which is called the "second signaling", is sent from the node to downstream nodes in order to reserve new wavelength channel for the buffered low-priority DB. The effects of network traffic load, the average length of DB, and the parameters of FDL buffer on DB's loss probability and end-to-end delay are researched by theoretic analysis and OBS network simulations. The results show that the proposed scheduling algorithm can obviously reduces the loss probability and end-to-end delay of high-priority DBs. So it can improve significantly QoS performances for high-priority DB in OBS networks.3. Differentiated Service Hybrid Signaling ProtocolIn order to support multiple traffics with differentiated services in OBS networks, especially real-time traffic, a novel differentiated service hybrid (DSH) signaling protocol is proposed. It tries to create an environment in which whether real-time services or non-real-time services can all be provided. It adopts tell and wait (TAW) protocol and just enough time (JET) protocol to deal with real-time traffics and non-real-time traffics respectively according to their QoS requirements. TAW signaling is modified to be adapted to transport real-time (or circuit emulation) traffics. A call request of the real-time traffic can be used to set up a virtual channel from a source node to a destination node before the traffic arrives, and the virtual channel can be held until the call end. Along the virtual channel a real-time traffic can be transmitted in a series of DBs with fixed size periodically. In this way, small end-to-end delay and low DB loss rate will be achieved for those accepted calls. DSH protocol supports multiple real-time traffics sharing one wavelength channel by TDM mode, which makes wavelength utilization higher and call loss probability lower.Theoretic analysis and computer simulations show that DSH protocol can implement differentiated services for real-time and no-real-time traffics, and ensure QoS performance for real-time traffic better. After comparing with OCS and other hybrid system, the results also show that the DSH protocol can make the lowest call loss probability for real-time traffics among them and significantly improve QoS performance of OBS networks.4. Feedback Congestion Control Scheme Employing Resource Management PacketsIn OBS networks, the one-way reservation signaling protocol results in DB contentions happening frequently at core nodes. Especially under the conditions of high traffic loads, more DBs will be lost and the network will be congested.A novel rate-based congestion control scheme for OBS networks is proposed in the paper, entitled resource management packet (RMP) feedback congestion control scheme. Edge nodes monitor congestion situations by sending and receiving resource management packets (RMP/RMPACK) in the control wavelength channel along the DB path periodically, which take the congestion information of core nodes in the OBS network, and adjust the sending rate of data bursts to avoid network congestion.. At the same time, the sources injecting IP packets to the edge nodes of OBS network can also be requested to change transmitting rate of IP packets by the high layer protocol.Performance of the proposed scheme is studied by theoretical analysis and computer simulation. Results show that RMP scheme can effectively control congestion and provide good differentiated services fairly for different traffics with different QoS requirements or through different light paths in OBS networks.