Chapter 1 Introduction
he wavelength-division multiplexer (WDM) has led to a tremendous increase in the available transmission capacity (e.g. 2.5, 10 and 40Gbps) for wide area networks (WAN). A traffic stream could be assigned to use one whole wavelength.
However, the required bandwidth of the traffic stream is usually much smaller than the capacity of a wavelength, and the available wavelengths are limited. Thus, many lower-speed traffic streams should be multiplexed onto a high-speed wavelength channel by traffic grooming to efficiently utilize the network resources. The number of low-rate traffic requests multiplexed in a wavelength channel is called the grooming factor. For example, if the bandwidth of a wavelength channel is OC-48 (i.e.2.488Gbit/s) and the base bandwidth of a connection is OC-12 (i.e.0.622Gbit/s), then four connections can be groomed and supported by one OC-48 channel. In this case, the grooming factor is four.
The problem of traffic grooming becomes increasingly important for emerging network technologies. An overview of the traffic grooming technique and survey of some typical works was reported in [1], [2].
T
In the design of traffic grooming, it can be classified into various categories, according to the network configuration, traffic and cost function. For the topology of networks, it can be categorized into ring or mesh. A ring network can also be a unidirectional ring or
bidirectional ring. For the traffic characteristics, according to the bandwidth request, it can be uniform or non-uniform. The bandwidth request of each user is the same, called uniform traffic; otherwise, the traffic is nonuniform. And, according to the system connection it also can be classified into dynamic traffic or static traffic. Static traffic means the traffic streams are set up all at once and fixed thereafter. Dynamic traffic means that traffic streams are set up and terminated at arbitrary time. For the objective cost function, traffic grooming may categorize into five types. The first one is the minimization of the number of Line Terminating Equipment (LTE). The second one is the reduction of the cost of actual electronic processing involved in Optical-Electric-Optical (O/E/O). The third one is to minimize the maximum number of lightpaths originating/terminating at a network node. The fourth one is the maximization of traffic throughput of the optical network. The last one is the minimization of blocking probability of new call request in dynamic traffic. Also, the traffic grooming can be classified into three categories: unicast (point-to-point), multicast (point-to-multipoint) and groupcast (multipoint-to-multipoint).
Many researches on traffic grooming in WDM networks have exclusively dealt with unicast traffic [3] - [8]. Static traffic was considered in the [3]-[6]. In [3] and [4], the authors considered uniform traffic in general topology network. The objective of the two papers is to minimize wavelength usage in the network. The authors formulated the problem into an integer linear program and proposed a simple approach to find a suboptimal solution. Non-uniform traffic in WDM ring was studied in [5] and [6]. In [5], the objective is to minimize total amount of electronic switching at all network nodes. The authors presented a new framework of bound which can be used to evaluate the performance heuristic. They gave both upper and lower bound which required less computation than the optimal solution.
In [6], both uni- and bi-direction ring were considered. The object of this paper was reducing both the number of wavelength and the number of LTE. Some lower bounds in
the number of wavelength and LTE required for a given traffic pattern were also derived.
The heuristic proposed in this paper can be considered in two phases. The first one is the circle formulation. Each circle consists of multiple non-overlapping connections. After the circles were constructed, optimal or near-optimal algorithms were used to groom circles onto a wavelength.
In [7] and [8], dynamic non-uniform traffic was considered in ring. In [7], a special topology of interconnected ring was considered, and the GA approach was used to find a topology with minimum number of ADMs to support a set of traffic matrices. Unlike the dynamic problem in [7], a dynamic provisions and grooming problem was considered in [8]. This kind of dynamic problem is measured by arrival time and holding time. The objective was to minimize the blocking probability.
Multicast is a kind of group communication, which requires simultaneous transmission of messages from a source to a group of destinations. It is expected that a sizable portion of the traffic in future high performance networks will be multicast, for example, multi-party conferencing, video distribution, network news distribution, and web content distribution to proxies. However, most multicast service applications require only sub wavelength capacity (for example, HDTV needs only 20Mbps). It would be inefficient to assign a whole lightpath to each lower-rate multicast streams. So the problem of multicast traffic grooming over optical networks has received significant attention [2].
Some multicast problems have been investigated in [9] - [13].A mesh network and the dynamic multicast traffic grooming were considered in [9] - [11]. The objective cost function is to minimize the loss probability. In [9], two simple approaches were proposed to groom dynamic traffic. The first approach is the single-hop (SH) traffic grooming. A new call has the same source and destination(s) with a working lightpath (or light tree) and the lightpath has enough bandwidth to support the new call, then, the new call will be groomed with the lightpath. The second approach is the multi-hop (MH) traffic grooming.
In MH, a lightpath (or light-tree) that has the same destination(s) with enough bandwidth which called “to destinations light-tree” (TDLT) was searched. After such a lightpath is found, another lightpath between the source of the new call and the source of TDLT by SH approach was selected. Note that only grooming by the SH, the traffic can be transmitted all-optically before reaching the destination(s).
An approach called hybrid provisioning grooming scheme was proposed in [10]. This approach improves the MH by combining the provision and un-provisioned lightpath and light-tree to serve a new connection. In MH, when a lightpath cannot be found between the source of a new call and the source of TDLT, the new call will be blocked. In this approach, the source would know if there is a TDLT to its destination(s). If enough resource was found, a new lightpath would be set up. In [11], the authors proposed multicast tree decomposition to improve the approaches used in [10]. The authors first divided a new multicast tree into many subtrees with one or two destination. Then, they tried to find multicast trees that have the same destinations with those subtree from current tree and groom them.
A ring topology with multicast static traffic was investigated in [12] and [13]. In [12], the authors considered single-source multicast traffic and the object is to minimize the number of LTE. An approximate algorithm was proposed by transforming the grooming problem into a weighted set cover problem. A bi-directional ring and uniform traffic was considered in [13]. The objective of this paper is to minimize the number of LTE. The authors proposed two heuristic algorithms. One is to use the minimum spanning tree routing and transform each multicast session to a set of unicast, and then, it uses the approach proposed in [6] to construct circles and groom circles. The other is to do the routing and circles construction simultaneously. The latter is effective than the former.
Ring networks have been studied in many researches. Most of them focus on unicast and static. Few dynamic and non-uniform multicast traffic researches ware considered in
ring network. Therefore, we are motivated to study the traffic grooming problem in optical bidirectional ring network with dynamic non-uniform multicast traffic. We propose the node architecture which is equipped in ring network and have the ability of traffic grooming and light splitting. Their object is to reduce the new call blocking probability and maximize the utilization of used wavelength.
The integer linear problem for the traffic grooming problem is formulated in mathematical form. However, the optimal solution by the ILP is infeasible due to the computation complexity. We propose a suboptimal multicast traffic grooming scheme, called Maximum Utilization and Minimum Hops (MUMO) scheme. Two main operations in MUMO: traffic routing and traffic grooming. In traffic routing, we propose a scheme which chooses the route with minimum hops for bidirectional ring and we choose the route according to the remained capacity in each fiber. So the routing scheme can save more resource for future calls. In the traffic grooming, we utilize the features of ring network and multicast traffic to effectively grooming the new call. The multicast traffic grooming scheme increases the probability of finding lightpath to groom and increases the utilization. Because of the two operations, MUMO can achieve the better utilization without changing the lightpath topology and reduce the blocking probability.