Non-greedy minimum interference routing algorithm for bandwidth-guaranteed flows
Introduction
One of the key aspects of future Internet is the ability to establish connections capable of meeting quality of service (QoS) requirements. The demand for such capability is caused by the rapid development in multimedia applications such as video-on-demand and video conferencing that require QoS guarantees, and the establishment of virtual private network (VPN) to satisfy customer service level agreements (SLAs). To deliver performance guarantees, reservation of resources is required. In particular, the capability to allocate resources in terms of bandwidth-guaranteed tunnels is essential in future Internet.
In reservation based networks, the resources (link bandwidth) which have been allotted to a particular connection are solely occupied and thus not available for other connections during the connection period. As a result, future connection set-up requests may be denied due to insufficient network resources. Therefore, there is a need for a routing and admission strategy that efficiently addresses the problems of selecting a path to route a request and deciding on whether the request should be routed or rejected. The obvious way to achieve this is by adopting offline algorithms, which require prior knowledge of point-to-point demands. However, due to the advent of services that permit dynamic and frequent requests for capacity change, online algorithms capable of handling requests that arrive one at a time without knowledge of future traffic arrival must be adopted. In addition, the algorithms must route the traffic without splitting it into multiple paths and the rerouting of existing connections should be minimized or avoided altogether.
A natural admission control scheme is a greedy strategy in which a request is accepted as long as an admissible path exists. However, the greedy strategy can lead to inefficient use of resources. For example, it will accept a connection request even if it spans excessively over a long path. Longer path ties up more resources and thus decreases network throughput. Consequently, much research has been focused on non-greedy routing and admission control strategies. The problem has been studied extensively in the context of circuit-switched networks. However, most of the analyses and design efforts were based on the assumption that the request arrival pattern is described by simple probabilistic model with known parameters. An example of such algorithms is the real time network routing algorithm (RTNR) [10] used in the AT&T long distance network. It is only recently that a new framework for routing strategies which do not rely on any assumption about the probabilistic behavior of the traffic was developed for general topology network [2], [3], [11]. Based on theoretic competitive analysis, these algorithms have been shown to exhibit good performance. In particular, a non-greedy algorithm called EXP [19] has been proposed based on the competitive algorithm developed in Ref. [2]. EXP has been shown to out-perform the greedy min-hop routing algorithm (MHA).
In light of the fact that the quasi-static knowledge on the ingress–egress points in the network should be exploited by routing algorithm to avoid potential connection rejection, Kodialam and Lakshman proposed the minimum interference routing algorithm (MIRA) [15] which utilizes such information in the context of multi-protocol label switching (MPLS) networks. MIRA was developed based on the ‘minimum interference’ concept that is a newly routed tunnel must follow a route that does not impart too much interference to paths of which availability may be critical in satisfying future demands. The simulation study in Ref. [15] shows that MIRA performs better than the min-hop routing and the widest shortest path algorithms [21].
In this paper, we present a new online non-greedy routing and admission control algorithm for the dynamic routing of bandwidth-guaranteed tunnels. As in Ref. [15], we will describe the algorithm in the context of MPLS networks for ease of terminology and conciseness. In MPLS networks, packets are assigned fixed-length labels at ingress routers which are then used to forward the packets along label switched paths (LSPs). The LSPs can be associated with certain constraints such as the bandwidth needed to satisfy customer SLAs. An important application of bandwidth-guaranteed LSPs is to become components of an IP VPN. Before mapping packets onto an LSP, the LSP is established using a signaling protocol such as the extension of resource reservation protocol (RSVP) [6] or label distribution protocol (LDP) [14]. A key feature of LSP is that it can be set up explicitly along a pre-specified path. This enables service providers to traffic-engineer their networks and to provide the bandwidth-guaranteed paths dynamically.
Our algorithm integrates the key concepts in the competitive strategies and minimal interference routing algorithm. Specifically, we introduce a new weight assignment scheme, which uses an exponential function to translate link's utilization and criticality into cost. The admission control scheme will accept the request if there is a sufficiently cheap path that is a path with the sum of the link costs less than a pre-defined threshold value.
We evaluate the performance of the new algorithm by carrying out an extensive set of simulations on a number of network topologies. The simulations show that our algorithm out-performs the original MIRA, min-hop routing and widest shortest path algorithms, and a variant of non-greedy scheme over a broad range of network operating environments. The simulations also show that our algorithm is capable of providing priorities to certain ingress–egress pairs in an environment which requires lower blocking probability for the tunnel set-up requests of the pairs.
The rest of this paper is structured as follows. In Section 2, a review of related work is presented. In Section 3, we will describe the system model and the problem definition. Reviews and analyses of the main concepts of minimal interference routing are presented in Section 4. In Section 5, our proposed algorithm is presented and analyzed. Discussion of simulation results and conclusion are presented in 6 Performance evaluation, 7 Conclusions, respectively.
Section snippets
Related work
In recent years, much research work has been carried out on various aspects of unicast QoS routing. For example, there has been research focused on path selection algorithms [2], [15], [19], [21], [23], [27], [28], [29], [30], cost and feasibility analysis [1], [8] and processing cost reduction strategies [9], [13], [16]. In this section, we describe some early works on unicast path selection algorithm based on different metrics. A complete overview of QoS routing problem can be found in Ref.
Problem definition and system model
The network is represented as a capacitated graph G(V,E,u) with n nodes and m edges, where u(e) represents the capacity of the edge e∈E. We consider the request for an LSP i as a triple (si, di, bi). Nodes si and di are the source and destination of the request and bi is the bandwidth required for the LSP. We assume that the requests for LSPs arrive one at a time and there is no knowledge of future demand characteristics.
We describe, for simplicity, only a centralized routing model in this
Review of minimum interference routing algorithm (MIRA)
In this section, we review the key concepts of MIRA [15]. The main concept of MIRA is to pick paths that do not interfere much with potential future LSP set-up requests of other source–destination pairs. In a capacitated network, the upper bound on the total amount of bandwidth that can be routed between a given ingress–egress pair is represented by the maximum flow (maxflow) value, v of the pair. The maxflow between the source node s and destination node d is the maximum flow value that can be
Non-greedy minimum interference routing algorithm
The idea of incorporating minimal interference in routing of MIRA has been shown to be effective by resulting in an algorithm which performs better than algorithms that do not consider the knowledge of ingress–egress points in the network. However, it has several limitations. Firstly, MIRA is a greedy algorithm. As the network resources are scarce, approach that attempts to admit all connection can result in inefficient resource utilization. Secondly, path selection in MIRA does not explicitly
Performance evaluation
In this section, we evaluate the performance of the E-MIRA and several other algorithms. The evaluation is carried out using an extensive set of simulation study. Although analytical models enable provable results be obtained, with the complexity of today's network topology, traffic distributions and service classes, it is extremely difficult to perform analytical modeling without sacrificing realism. In our simulation study, we choose a diverse range of network topologies and traffic patterns.
Conclusions
The primary contribution of the paper is the development of a non-greedy routing and admission control algorithm for the dynamic routing of bandwidth-guaranteed tunnels. The algorithm combines the advantages of algorithms which were developed based on the notion of minimum interference and competitive analysis. It is based on the objectives to minimize interference under low network loading and resource usage under high network loading. In addition, paths which consist of links that cause large
References (29)
- et al.
Network path caching: issues, algorithms and a simulation study
J. Comput. Commun.
(1997) - et al.
A randomized algorithm for finding a path subject to multiple QoS constraints
J. Comput. Networks
(2001) - et al.
Evaluating the overheads of source-directed quality-of-service routing
Proc. IEEE Int. Conf. Network Protocols (ICNP'98), Austin TX, October
(1998) - et al.
Throughput competitive on-line routing
Proc. 34th Annual Symp. Found. Comput. Sci., Polo Alto, California, November
(1993) - et al.
Competitive routing of virtual circuits with unknown duration
Proc. Fifth ACM-SIAM Symp. Discrete Algorithms, January
(1994) - D. Katz, D. Yeung, Traffic engineering extensions to OSPF, Internet Draft, August...
- et al.
QoS routing in networks with uncertain parameters
IEEE/ACM Trans. Network.
(1998) - D.O. Awduche, L. Berger, D. Gan, T. Li, G. Swallow, V. Srinivasan, RSVP-TE: extensions to RSVP for LSP tunnels,...
- et al.
How to model an internetwork
Proc. IEEE INFOCOM, March
(1996) - et al.
Intra-domain QoS routing in IP networks: a feasibility and cost/benefit analysis
IEEE Network (Spec. Issue Integ. Different. Serv. Internet)
(1999)
Server based QoS routing
Proc. IEEE GLOBECOM'99, Rio de Janeiro, Brazil
Real-time network routing in the AT&T network-improved service quality at lower cost
Proc. IEEE GLOBECOM, Orlando, Florida, December
On-line load balancing with applications to machine scheduling and virtual circuit routing
Proc. 23rd Annual ACM Symp. Theory Comput., San Diego, California, May
Cited by (4)
Study of minimum interference routing algorithm for MPLS traffic engineering
2006, Ruan Jian Xue Bao/Journal of SoftwareA distributed algorithm for inter-domain resources provisioning
2006, 2006 2nd Conference on Next Generation Internet Design and Engineering, NGI 2006A new hybrid traffic engineering routing algorithm for bandwidth guaranteed traffic
2006, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)A precomputation scheme for Minimum Interference Routing: The Least-Critical-Path-First algorithm
2005, Proceedings - IEEE INFOCOM