Over-Constraint QoS Routing in Large Networks W. S. Goodridge† W. R. Robertson†W. P. Phillips† and Shyamala Sivakumar‡ † Dalhousie University, ‡ St. Mary's University Abstract: Routing using more than one metric is difficult. Distributed Multimedia applications have stringent QoS demands on the network. However, many current QoS routing algorithms have large execution times when strict (small) user constraints are used. We present an algorithm that have relatively small execution times under strict constraints and demonstrate via simulation that this is so. 1 Introduction. Real-time network applications like telemedicine, video conferencing, and virtual reality will demand highest quality of service (QoS) and reliability from the data transport infrastructure. The networking community has developed frameworks such as InServ and Diffserv to alter performance for network flows by changing the schedule in a router or switch. However, there is another way to achieve QoS, which is called QoS Routing and the goal is to select a feasible path for a network flow based on the requirements of the flow. A lot of research has been done on QoS Routing and three main challenges are evident: firstly, finding a feasible path is a very complex problem; secondly, the routing protocols will have to measure and exchange values for multiple network metrics such as delay, jitter and bandwidth; and thirdly, a mechanism for the user or application to communicate a set of user constraints to the network is needed. In this paper we will only consider the challenge of finding feasible paths in communication networks. Finding a path subject to multiple metrics is difficult and is considered NP-complete [1]. Since the motivation for QoS Routing is rooted in the fact that multimedia applications require stringent user constraints, a good question is whether or not existing QoS routing algorithms can address the needs of applications that have strict constraints (over- constraints [2]). In this paper we will give a background on some of the QoS routing algorithms in section 2, and in Section 3 answer this important question. In Section 4, we introduce an alternative QoS Routing algorithm specially designed to work in over-constraint environments. In Section 5, we present simulations to support the claim that our algorithm works well in large networks. 2. Existing QoS Routing Algorithms Before we get into existing QoS algorithms we will give a formal definition for two problems associated with QoS Routing. Consider a graph G(N;E) consisting of a set of additive metrics w i (e) for each link e ∈ E, and user requested constraints L i , i ∈ [1, m]. The goal of multi-path-constrained routing (MCP) is to find a path P from source node s to destination node t such that w i (P) ≤ L i for all i. A path that satisfies all m constraints is often referred to as a feasible path. There may be multiple paths in the graph G(N; E) that satisfy the constraints, any of which are solutions to the MCP problem. Hence it may be necessary to use some type of optimization criteria to select a path from the set of feasible paths. This more difficult problem is called the multi-constrained optimal path (MCOP) problem. Because both the MCP and MCOP problems are NP-complete the bulk of the solutions proposed are heuristics. These heuristics algorithms include Jaffe [3], Chen [4], Iwata [5], and TAMCRA [6]. Typically these algorithms are highly specialized and cannot adapt easily to a wide range of user requirements. There also exist exact solutions for the MCP/MCOP like the SAMCRA [6] and H_MCOP [7]. These algorithms offer good performance at the expense of possible high complexities and running times growing exponentially in the worst case. However, in [8] it is argued that in practice an exact algorithm may work in polynomial time, making guaranteed QoS possible.