Contents I. Introduction
Determining the route taken by real time Internet applications, like voice over IP [1] forms the most dealt problem recently in network resource management process since it involves resource provisioning decision on the scale of the entire network. Quality of Service (QOS) based routing is of critical importance in achieving such applications with high speed and efficiency. QOS routing problem is involved in several constraints, such as, delay, delay jitter, packet loss rate, bandwidth and cost. QOS measures can be roughly classified into additive, multiplicative and concave. Bandwidth is an example of concave metric, whereas delay, cost, etc. are additive metrics. In case of an additive measure, the QOS value of a path is equal to the sum of the corresponding weights of the links along that path. For a non additive measure like bandwidth, the QOS value of a path is the minimum link weight along that path. In general, non additive measures can easily be dealt with by pruning from the graph all links that do not satisfy the requested QOS constraint [2]. Additive measures cause more difficulties. If we consider all QOS constraints in routing algorithm design, the corresponding algorithm will be certainly too complicated to be applicable. For this reason, researches in network resource management simplify QOS constraints and present a QOS routing model based on bandwidth guarantees. Least of them has been done on laying paths with both bandwidth and delay constraint. Considering only QOS constraints in resolving routing problem is not enough to satisfy real time Internet applications. We should take into account some Traffic Engineering (TE) objectives leading to the usage efficiency of network infrastructure, such distributing network load. Much of the researches in network have tried to provide bandwidth guarantees and preventing network congestion but least of them have been done on path selection algorithm using bandwidth, delay, and cost criteria. Traditional MIN-DELAY (Minimum Delay) [3] is a bandwidth-delay based routing that prunes links that do not satisfy the bandwidth requirement, and then finds the shortest path on the basis of delay metric in the reduced graph. Although simple and efficient, MIN-DELAY algorithm can create bottlenecks for future LSPs, and leads to network under-utilization. In [4] Blanchy, Melon, and Leduc present an online traffic engineering algorithm, called DAMOTE, to balance the load in network on the basis of a given objective function. DAMOTE is a bandwidth constrained routing algorithm which protects traffic with strong delay requirement by a rapid restoration in case of failure. In this approach, DAMOTE protects each primary path by a series of detour LSPs. These LSPs have to be pre-established and provisioned with bandwidth resources for fast restoration in case of failure. The drawback of DAMOTE, it concentrates on load balancing and does not consider end-to-end delay requirement. Reducing blocking of future requests is also an important TE objective. It was considered by authors in [5] and [6]. In [5], authors present the MIRA algorithm. They are based on the intuition that the blocking probability of requests can be reduced if the total available flow in the network is maximized. To maximize network flow, they rely on the fact that each newly routed connection should follow a path that does not interfere with a path that may be critical to satisfy future demands. Similar to MIRA, NEWMIRA has been proposed in [6] to overcome some MIRA limitations by taking into account the overall bandwidth blocking effects of routing an LSP request. Both MIRA and NEWMIRA do not take into account delay and cost constraints.
