Rerouting tunnels for MPLS network resource optimization

Abstract

In Multi-Protocol Label Switching (MPLS) networks, traffic demands can be routed along tunnels called Label Switched Paths (LSPs). A tunnel is characterized by a path in the network and a reserved bandwidth. These tunnels can be created and deleted dynamically, depending on traffic demand arrivals or departures. After several operations of this type, the network resource utilization can be unsatisfactory, with congestion or too long routing paths for instance. One way to improve it is to reroute tunnels; the rerouting process depends on the LSP Quality of Service (QoS) requirements.

Three levels of QoS are considered, with three associated types of LSPs. A global rerouting framework is proposed, which enables us to consider independently each type of LSP. Then, mathematical models are introduced and analyzed. A focus is made on complexity analysis and optimal resolution of these problems. Finally, some numerical results illustrate the theoretical analysis.

Introduction

In Multi-Protocol Label Switching (MPLS) networks, traffic demands can be routed along tunnels called Label Switched Paths (LSPs) [2], [3]. An LSP corresponds to a path in the network with bandwidth reserved. Depending on traffic demand arrivals, LSPs can be dynamically created to route them. Similarly, traffic demand departures can lead to removing some LSPs from the network. After several creations/deletions of LSPs, the network resource utilization can become very unsatisfactory. This can hardly be avoided, since future events are not known when establishing an LSP.

A way to improve the situation at a given time is to reroute the existing LSPs into a better global configuration. This rerouting process is performed “off-line” during a quiet period when the network state is stable. Different levels of quality of service (QoS) have to be considered, depending on the services supported by an LSP. Thus, the rerouting plan has to take into account these QoS differences. Indeed, there exist different ways of rerouting LSPs, which have different effects on the traffic. Three different classes of service are considered in this paper, with high, medium and low QoS. Low quality LSPs can be broken and re-established afterwards. Thus, the service is unavailable for possibly a few seconds. A medium level corresponds to the possibility of rerouting the LSPs, but through a “make-before-break” process (see [2]): first, a new LSP is established, and then, the old one can be deleted. In practice, this process has little impact on the communication quality, and can induce some packet losses. The highest QoS level requires that the corresponding LSPs cannot be moved at all.

In [9], [10], [11], the authors have proposed a first approach to this rerouting issue, considering only medium class LSPs. They study the problem the following way: knowing the current (old) configuration and the optimal (new) one, they look for a feasible rerouting sequence. Each LSP is rerouted only once, i.e. no intermediate path is used. This approach emphasizes the optimality of the final routing which is calculated from scratch. However, it may be necessary to break a few connections to be able to reach this target. Conditions on capacities in the network are given to ensure the existence of a rerouting sequence without connection breaking. This previous work has not taken into account the different classes of service which could exist in such a network. Moreover, even when considering only medium class LSPs, the proposed method is not totally satisfactory. On the one hand, the LSPs are allowed to be possibly broken. On the other hand, the number of reroutings to perform is possibly equal to the number of LSPs in the network. This number can be large, implying some complexity in the network management.

In the current paper, a complementary approach is proposed. The emphasis is put on fulfilling the different levels of quality of service. In particular, medium quality LSPs cannot be broken. With this constraint, we try to obtain the best possible state. Moreover, the maximum number of reroutings can be controlled, the goal being to keep network management as easy as possible and to minimize the service disturbances. The problem addressed has many connections with that of [1], where a rerouting problem is studied with the aim to improve a telecommunication network state. The authors assume that each path in the network is assigned a usage cost, and heuristics are designed to find a rerouting sequence which leads to a small global network cost. But such a fixed path cost can hardly model QoS issues such as those considered in the current study. Moreover, we focus on exact solution procedures to find optimal solutions.

Finally, such rerouting problems occur in fields other than telecommunications. The recent work of [17] deals with moving processes from their initial processor to another one in order to improve the computation resource utilization. This problem is in fact a special case of that exposed in [9], [10], [11]. This paper presents in particular a good review of the related literature, which shows that similar problems have in fact quite rarely been studied in the past.

In Section 2, a global rerouting framework is proposed, which enables us to consider independently the different classes of QoS. Section 3 deals with medium quality LSP rerouting, while low quality LSPs are studied in Section 4. In both sections, mathematical models are established and optimal resolution is more particularly investigated. Section 5 provides some numerical results.

Note that only point-to-point (P2P) LSPs are dealt with; the problem is more difficult for point-to-multi-points ones (P2MP). Finally, in MPLS networks, several traffic demands can use the same LSP. Nevertheless, from now on, without loss of generality for our study, the words “demand”, “tunnel” and “LSP” will denote the same thing.