Quality of Service Routing in MPLS Networks Using Delay and Bandwidth Constraints

Quality of Service Routing in MPLS Networks Using Delay and Bandwidth Constraints Mohammad HossienYaghmae Computer Department, Faculty of Engineering, Ferdowsi University of Mashad, Mashhad, Iran hyaghmae@ferdowsi.um.ac.ir and Ali Asghar Safaeei Computer Department, Faculty of Engineering, Ferdowsi University of Mashad, Mashhad, Iran Al_Sa69@stu-mail.um.ac.ir Abstract Multi-Protocol Label Switching (MPLS) has been proposed as a new approach for integrating layer 3 routing with the layer 2 switching. It integrates the label swapping paradigm of layer 2 (e.g. ATM and Frame Relay) with the routing of layer 3 (e.g. IP and IPX). In the MPLS networks, constraint-based routing computes routes that are subject to constraints such as bandwidth and administrative policy. Because constraint based routing considers more than network topology in computing routes, it may find a longer but lightly loaded path better than the heavily loaded shortest path. In this paper we propose a new constraint based routing algorithm for MPLS networks. The proposed algorithm which is a modification of [] algorithm, uses both bandwidth and delay constraints. It means that the reservable bandwidth of all of the links along computed path must be equal to or greater than the bandwidth constraint value and the delay of the path must be less than or equal to the delay constraint value. In the proposed algorithm, the best path is selected based on strategy. Using the MPLS Network Simulator ( MNS)[2], the performance of the proposed algorithm is compared with that of [] and [3] algorithms. Simulation results show that in comparison with the other methods, the proposed algorithm has a better performance. Keywords: MPLS networks, Quality of Services (QoS), Constraint Based Routing, Routing Protocols, Traffic Engineering. Introduction The Internet is rapidly grown. During past few years, the demand for using multimedia applications over the Internet has been increased. By developing new Internet applications, it is a problem how to satisfy the quality of service (QoS) requirements of these new applications, e.g., requirements regarding bandwidth, delay, jitter, packet loss, and reliability. The Internet Engineering Task Force (IETF) has proposed three different service models and mechanisms to support the requested QoS. These models are: Integrated Services (Intserv)[4]/ Resource Reservation Protocol (RSVP) [5-6], Differentiated Services (Diffserv) [7-8] and Multi-Protocol Label Switching (MPLS) [9-], The MPLS can provide fast packet forwarding and traffic engineering. It is a new Internet technology that is rapidly emerging as a core technology for next-generation networks, both traditional data networks and increasingly optical networks. The efficiencies and new capabilities enabled by MPLS make it ideal for cable operators expanding their broadband offerings. MPLS is designed for use from edge-to-edge of converged data and voice networks. Rather than replace IP routing, MPLS is designed to overlay its functionality on top of existing and future routing technologies and to work across a variety of physical layers to enable efficient data forwarding together with reservation of bandwidth for traffic flows with differing QoS requirements. MPLS uses a technique known as label switching to forward data through the network. A Label Edge Router (LER) inserts a small, fixed-format label in front of each data packet once it enters into the MPLS network. This packet is then handed off to a Label Switch Router (LSR) as it traverses This research was in part supported by a grant from Iran Telecommunication Research Center (ITRC) P54/

through the core of the network. At each hop across the network, the packet is routed based on the value of the incoming interface and label, and dispatched to an outwards interface with a new label value. The path that data follows through a network is defined by the transition in label values, as the label is swapped at each LSR. Since the mapping between labels is constant at each LSR, the path is determined by the initial label value. This path is called Label Switched Path (LSP). MPLS offers a host of benefits, chief among them is network efficiency. It provides more efficiencies across networks regardless of other platforms for several reasons. MPLS can be mapped across various media, including Ethernet, ATM and SONET. In addition, MPLS can operate on top of various routing protocols, including OSPF, RIP and BGP. In order to minimize the congestion occurred by unfair load balancing, Traffic Engineering (TE) is performed to maximize total throughput of network. Traffic Engineering is the first application of MPLS networks [-2]. Although IP networks offer flexibility and scalability, the existing IP networks need to be enhanced in the areas of availability, dependability, and quality of service, in order to provide a missioncritical networking environment. Traffic engineering is a critical building block needed to reach this goal. Currently, IP traffic is routed and forwarded over the Internet using standard IP routing protocols. Interior Gateway Protocols (IGPs, for example, OSPF and RIP) are used within Autonomous Systems (ASs), while Exterior Gateway Protocols (EGPs, such as BGP4) are used to interconnect ASs. IGPs and EGPs offer little or no readily available traffic control capabilities, such as reporting and incorporating in routing the information of network resource availability or utilization. The routing algorithms (e.g. shortest path first or distance vector) used by routing protocols tend to converge traffic onto the same network links or interfaces, which contributes significantly to congestion and unbalanced network. In the MPLS networks, to set up a hop-by-hop LSP, the Label Distribution Protocol (LDP) has been proposed. One new technique proposed to implement explicit routing is the Constraint based Routing using Label Distribution Protocol (). builds upon LDP protocol, which is already part of MPLS. On the other hand, Differentiated Services (DiffServ) is a scalable class of service architecture proposed by the IETF, which provides scalable QoS guarantee. The Diffserv networks can support different service models including the Expedited Forwarding (EF), the Assured Forwarding (AF) And The Best Effort (BE). The AF provides for the delivery of IP packets in four independent delivery classes (AF to AF4), where each class is allocated a certain amount of resources, such as buffers and bandwidth, in each DiffServ node. By integrating the DiffServ and MPLS, as defined in [3], a very attractive strategy to backbone networks service providers with scalable QoS and traffic engineering capabilities can be obtained. The DiffServ provides scalable edge-to-edge QoS, while MPLS performs traffic engineering to evenly distribute traffic load on available links and fast rerouting to route around node and link failures. One of the key issues in providing QoS guarantees is how to determine paths that satisfy QoS constraints. Solving this problem is known as QoS routing or constraint-based routing. The research community has extensively studied the QoS routing problem, resulting in many QoS routing algorithms. Routing in general involves two entities, namely the routing protocol and the routing algorithm. The routing protocol manages the dynamics of the routing process: capturing the state of the network and its available network resources and distributing this information throughout the network. The routing algorithm uses this information to compute paths that optimize a criterion and/or obey constraints. The traditional routing algorithms use one metric such as hop count or cost and forward all of the traffics by the shortest path. So this path would be congested and TE must be done. Thus, one of the TE components is QoS routing algorithm [4], which considers more QoS metrics as its constraints and finds the best path, which satisfies these constraints. In [5] it has been shown that when a class of WFQ-like scheduling algorithm is used, the end-to-end delay, the delay-jitter, and the buffer space bounds are not independent. They are functions of the reserved bandwidth, the selected path and the traffic characteristics. Therefore, the problem of finding a path satisfying the bandwidth, the delay, the delay-jitter and the buffer space constraints, which is NP-complete in general, can be solved by a modified version of Bellman- Ford algorithm in polynomial time. In [6] a heuristic algorithm for the NP-complete multi-path-constrained routing problem has been proposed. This reduces the original NP-complete problem to a simpler problem solvable in the polynomial time. In [7] a new online algorithm for dynamically routing bandwidth guaranteed label switched paths has been introduced. Other LSP path QoS metrics, such as delay and losses, are not explicitly considered in this work. With attention to this summarized review, each one of QoS routing algorithms solves a specific routing problem using different constraints, special suppositions and circumstances. The nearest QoS routing algorithm to our proposed algorithm is the [] algorithm. In [] by using the Dijkstra s shortest path algorithm, a QoS routing algorithm that finds a path that satisfies two constraints bandwidth and delay, has been proposed. P54/2

The remainder of the paper is organized as follow. In section 2,we explain the proposed algorithm in details. In section 3, by using the MPLS Network Simulator (MNS), we evaluate the performance of proposed algorithm with those of and algorithms. Finally, section 4 concludes the paper. 2. Proposed Algorithm Based on the previous studies, it can be shown that any two or more combinations of two additive metrics are NP-complete. As shown in [] the only feasible combinations are bandwidth with one of four (delay, delay jitter, cost, packet loss probability). As the bandwidth and propagation delay are two important constraints, in the proposed algorithm we use only bandwidth and delay constraints. We can make sure that these two metrics are not inter-dependent and our routing problem won t be NP-complete. As the MPLS networks use the explicit routes where the path must be computed in the source router so, the routing strategy of our proposed algorithm is the source routing. Furthermore, it is a unicast routing algorithm. The approach used in design of the proposed algorithm, first solves the QoS routing problem, then performs optimization and finally considers the future requests. In order to solve the problem, it must find a feasible path that satisfies both bandwidth and delay constraints. It means that the reservable bandwidth of all of the links along computed path must be equal to or greater than the bandwidth constraint value and the delay of the path must be less than or equal to the delay constraint value. After we make sure that the computed path satisfies the requirements, if there are more than one feasible path, we must choose the best one. The best path is the shortest one which has the least delay value. In order to optimize network resource utilization, the path with minimum of hop count is preferred. The other optimization aspect is related to computation time complexity. The order of a routing algorithm is tightly dependent on the number of nodes / links in the network topology graph. Therefore if the dimension of this graph could be reduced, the path computation time complexity will be better. To achieve this goal, we prune the topology graph as well as we are able to. In order to minimize the probability of congestion occurrence for the future requests, the load-balancing rule is used. To conserve large capacity of the links (large amount of bandwidth) for future requests, which their required bandwidth value is large, the strategy in the load-balancing step is preferred. The high level description of our proposed QoS routing algorithm is summarized as bellow: ) Prune the topology graph from these links: Suppose that (i,j) is the link between nodes i and j. a) if Bandwidth(i,j) < Bandwidth constraint, then delete (i,j) from topology graph. b) if Delay(i,j) > Delay constraint then delete (i,j) from topology graph c) if (Delay(i,j)==Delay Constraint) &&!(i == source node && j== destination node) then delete(i,j) from the topology graph d) if node i is of degree (i has only one neighborhood, j ) &&!((i == source node && j == destination node) (i == destination node && j == source node)) then delete (i,j) from the topology graph. 2) Assign link s delay to their cost and find the shortest path using Dijkstra s algorithm. 3) if there are no feasible path, reject the request, but if there are more than one feasible path, select the path with the minimum hop count. 4) if more than one path are remained from step 3, the load-balancing rule is performed. In the proposed algorithm, the approach which is used in memory management of operating systems, has been chosen. If there are several requests at a time, this algorithm process them in a decrement order of their bandwidth constraint values. In step, the algorithm eliminates all links with bandwidth value less than the bandwidth constraint. So, it is clear that all remained links will satisfy the bandwidth constraint. The other components of this step help to have a better time complexity. In step 2, it finds the path with delay value less than (or equal to) delay constraint value (if there is any). So, up to here, the desired QoS routing problem is solved and the two constraints are satisfied. But in this step the best path between all feasible ones is selected (optimization). In step 3, if the path with minimum hop count is used, minimum of network resources (such as buffer space, CPU time, ) are consumed. In step 4, the load is distributed and balanced. So, the congestion occurrence probability for future requests will be reduced. The loadbalancing strategy selects a path from all feasible paths which has the nearest reserveable / residual bandwidth value to the bandwidth constraint value. P54/3

3. Performance Evaluation In this section, by using MPLS Network Simulator (MNS) [2], the performance of our proposed QoS routing algorithm is compared with those of and algorithms. The evaluation parameters are the throughput, the call blocking rate, the path s delay and the end-to-end delay. The traffic parameters of AF, EF and BE services are shown in table. Table. Traffic parameters used in the simulation Service Model Bandwidth Traffic type Mean burst time Mean silence time AF 8 kb/s On/Off 5 ms 5 ms AF2 kb/s Constant Bit Rate - - EF 8 kb/s Constant Bit Rate - - BE kb/s On/Off 2 ms 8 ms The call requests are reached at each.5 s, periodically. The network topology used in the simulation is shown in figure. In this figure the numbers a, b shown in each link, represent the number of requests of the desired service class which can be transmitted by this link and the link s propagation delay (in ms), respectively. In figures 2 for all traffic classes AF,AF2, EF and BE, the throughput is plotted versus simulation time. It can easily be seen that the proposed algorithm has better performance than those of and algorithms. S, 2,2 3,2 5,4 2 2,2 2,2 2,3 3,3 5 3,, 3 6, 4,4,3 4 8 3,2 2,2 6,2 9 3,2 2,3 7, D Figure. Network topology used for simulation 7. 6.5 6. 5.5 5. 4.5 4. 3.5 3. 2.5 2..5..5. Throughput versus time (EF Traffic)..29 2.29.79.9 2.37 2.87 3.45 3.92 4.29 4.45 4.79 5. 9 8 7 6 5 4 3 2 Throughput versus time (AF2 traffic).585.59 2.86 2.28 2.667 3.68 3.745 4.245 4.59 4.746 5.22 5.59 5.82 6.68 8 Throughput versus time (AF traffic) Throughput versus time (BE traffic) 3.9.8.7 2.6.5.4.3.2. 3.7 3.2 3.34 3.87 3.96 4.66 5.5 5.25 5.4 5.89 6.5 6.7 7.4 7.27 8. 4.559 4.94 4.964 5.4 5.639 6.49 6.674 7.49 7.244 7.879 Figure 2. Throughput versus simulation time a) EF traffic b) AF2 traffic c)af traffic d)be traffic P54/4

In figures 3, for all traffic classes AF,AF2,EF and BE, the throughput is plotted versus the traffic load. This figure show the average throughput of each class of service based on the request arrival and regardless of the time. So, it doesn t indicate the momentary difference of throughput but it shows the total throughput of each service class regard to received request. Based on results shown in this figure, for all traffic types, the proposed algorithm has better throughput Throughput versus traffic load (EF traffic) Throughput versus traffic load (AF 2 traffic) 7. 6.5 6. 5.5 5. 4.5 4. 3.5 3. 2.5 2..5..5. Traffic Load (Mb/s)..8.6 2.4 3.2 4. 4.8 5.6 6.4 7.2 8. 8... 9. 8. 7. 6. 5. 4. 3. 2..... 2. 3. 4. 5. 6. 7. 8. 9... 3. 2.5 2..5..5. Throughput versus traffic load (AF traffic)..3.6.9.2.5.8 2. 2.4 2.7 3. 3...5. Throughput versus traffic load (BE traffic)...2.3.4.5.6.7.8.9.. Figure 3. Throughput versus traffic load a) EF traffic b) AF2 traffic c)af traffic d)be traffic In the next simulation trial, we evaluate the call-blocking rate of the proposed algorithm. The call-blocking rate is defined as bellow: number of rejected calls call blocking rate = total number of recieved calls In figure 4, the call-blocking rate of all algorithms is plotted versus traffic load. It is shown that for all types of traffic, our proposed algorithm has a better performance than the algorithm. Furthermore it is observed that algorithm has the same performance like the proposed algorithm, but note that it can not satisfies the delay constraints while our proposed algorithm can does it. This will be shown in the next figures. As mentioned before, the proposed algorithm considers only two constraints including bandwidth and delay. Based on results shown in the figures 2-4, in comparison with the and algorithms, the proposed algorithm has a better throughput and call blocking rate. In the next trials, we evaluate the delay performance of our proposed algorithm. For this purpose, we consider a new MPLS network topology shown in figure 5. We simulated this topology in MNS simulator. The numbers a,b shown in figure 5, represent the link capacity (in Mb/s) and the link s propagation delay (in ms),respectively. To compare the delay performance of the proposed algorithm with that of algorithm, we performed new simulation trials. In this case we considered four different scenarios. In each scenario, 8 consequences call request arrive at each time unit. Table 2 shows the requested bandwidth of this scenario. In figure 6, for both the proposed algorithm and algorithm, the delay of selected path is plotted versus traffic load. This figure shows that in comparison with the algorithm, the proposed algorithm can select a better path with minimal delay. P54/5

Call blocking rate versus traffic load (EF traffic) Call blocking rate versus traffic load (AF 2 traffic) 5% 3% 4% 3% 2% 2% % % %.8.6 2.4 3.2 4 4.8 5.6 6.4 7.2 8 8 % 2 3 4 5 6 7 8 9 Call blocking rate versus traffic load (AF traffic) Call blocking rate versus traffic load (BE traffic) 4% 5% 3% 4% 2% % 3% 2% % % %.3.6.9.2.5.8 2. 2.4 2.7 3 3..2.3.4.5.6.7.8.9 Figure 4. Call blocking rate versus traffic load a) EF traffic b) AF2 traffic c)af traffic d)be traffic 2.3,, 2.3, LSR2 LSR5 LSR8 LSR 2.3,.3,.3, 2.3, N 4,.3,.7,.3, LSR LSR4 LSR7 LSR9 4, N.3,.3, LSR3.3, LSR6 Figure 5. A MPLS network topology used in the simulation Call No. Table 2. Requested bandwidth of four different scenarios Requested Requested bandwidth bandwidth (scenario 2) (scenario 3) Requested bandwidth (scenario ) Requested bandwidth (scenario 4) 5 Kb/s 5 Kb/s 4 Kb/s 5 Kb/s 2 5 Kb/s 5 Kb/s 4 Kb/s 5 Kb/s 3 5 Kb/s 4 Kb/s 5 Kb/s 4 Kb/s 4 5 Kb/s 5 Kb/s 5 Kb/s 5 Kb/s 5 4 Kb/s 4 Kb/s 5 Kb/s 4 Kb/s 6 4 Kb/s 5 Kb/s 5 Kb/s 4 Kb/s 7 5 Kb/s 5 Kb/s 5 Kb/s 5 Kb/s 8 5 Kb/s 5 Kb/s 5 Kb/s 5 Kb/s P54/6

Path's delay versus traffic load Path's delay versus traffic load 2 2 Delay (ms) 9 8 7 6 5 4 3 2 Delay (ms) 9 8 7 6 5 4 3 2 5 25 4 8 2 7 22 5 55 95 5 2 25 22 Traffic load (Kb/s) Traffic load (Kb/s) Path's delay versus traffic load Path's delay versus traffic load 2 2 Delay (ms) 9 8 7 6 5 4 3 2 Delay (ms) 9 8 7 6 5 4 3 2 4 8 3 8 95 2 25 22 5 4 8 95 2 25 22 Traffic load (Kb/s) Traffic load (Kb/s) Figure 6. The path s delay versus traffic load a) scenario b)scenario 2 c)scenario 3 d)scenario 4 In figure 7, for all traffic classes, the end-to-end delay of the proposed algorithm is compared with those of and algorithms. In this figure it can be seen that for all types of traffic, our proposed algorithm has the better performance than the and algorithms. End-to-End delay versus traffic load (EF traffic) End-to-End delay versus traffic load (AF 2 traffic) End-to-End delay (s).25.24.23.22.2.2.9.8.7.6.5.4.3.2...8.6 2.4 3.2 4 4.8 5.6 6.4 7.2 8 End-to-End delay (s).56.55.54.53.52.5.5.49.48.47.46.45.44.43.42.4.4 2 3 4 5 6 7 8 9 End-to-End delay versus traffic load (AF traffic) End-to-End delay versus traffic load (BE traffic).26 2.3.24.22 2.28 End-to-End delay (s).2.8.6.4.2..8.6 End-to-End delay(s) 2.25 2.22 2.9 2.6.4.2 2.3 2..3.6.9.2.5.8 2. 2.4 2.7 3..2.3.4.5.6.7.8.9 Figure 7. End-to-End delay versus traffic load a) EF traffic b) AF2 c)af traffic d)be traffic In the next trial, different call request processing orders are compared with each others. We consider four scenarios including: )increment bandwidth order (Incr BW Order), 2)decrement bandwidth order (Decr BW Order), 3)service order (Service Order) and 4)random order (Random Order). In the increment bandwidth order, the incoming call which needs the least bandwidth has been selected, while in the decrement bandwidth order, the incoming call which needs the highest bandwidth has been selected. In the service class P54/7

order, the incoming requests are processed based on their requested class of service. Finally, in the random order style, the incoming call requests are processed randomly. The results are shown in figure 8,9. In this trial a combination of traffic sources including: AF,AF2, EF and BE are applied to the network topology shown in figure 4. In figure 8, the throughput is plotted versus simulation time. It can be seen that the decrement bandwidth order scenario has the highest throughput 4 3 2 9 8 7 6 5 4 3 2 Throughput versus time Incr BW Order Decr BW Order Service Order Random Order 2 2.33 2.4 2.5 2.66 2.69 2.74 2.83 2.9 3 3.2 3.8 3.29 3.49 3.53 3.72 3.79 3.93 3.98 4.2 4.6. Figure 8. The study of call request processing order: throughput versus simulation time In figure 9, the call blocking rate is plotted versus number of incoming requests. It can be seen that by increasing the number of requests, for all different call request processing orders, the blocking rate is increased. The results shown in figure confirm that in comparison with the other methods, the decrement bandwidth order has the better performance. 32% 28% Call blocking rate versus number of requests 24% 2% 6% 2% 8% Incr. BW Order Decr.BW Order Service Class Random Order 4% % 2 3 4 5 6 7 8 9 2 3 4 5 6 7 8 9 2 Number of requests Figure 9. The study of call request processing order: call blocking rate versus number of requests 4.Conclusion In the MPLS networks, by using constraint based routing and enhanced IGP, traffic engineering can be done much more effectively. The traditional routing algorithms use one metric such as hop count or cost and forward all of the traffics by the shortest path. So this path would be congested and traffic engineering must be done. Once a suitable QoS routing protocol is available and each node in the network has an up to date view of the network, the challenging task in QoS routing is to find a path subject to multiple constraints. Each one of QoS routing algorithms solves a specific routing problem using different constraints, special suppositions and circumstances. In this paper we presented a constraint based routing algorithm for the MPLS networks. In the proposed algorithm, the approach which is used in memory management of operating systems, has been chosen. If there are several requests at a time, this algorithm process them in a decrement order of their bandwidth constraint values. By using the MNS simulator, the performance of our proposed QoS routing algorithm was evaluated and compared with those of and algorithms. The simulation results easily helped us to judge the merit of the proposed algorithm because of its remarkable performance over the other algorithms. P54/8

References [] Z. Wang and J. Crowcroft, QoS Routing for Supporting Multimeda Application, IEEE Journal on Selected Areasin Communications. [2] MPLS Network Simulator (MNS), http://flower.ce.cnu.ac.kr/~fog/mns/ [3] B. Jamoussi, Ed., et. al. Constraint-Based LSP Setup using LDP, IETF RFC 322, January 22. [4] R. Braden, D. Clark, S. Shenker, Integrated Services in the Internet Architecture: an Overview, IETF RFC 633, June 994. [5] R. Braden, Ed., L. Zhang, S. Berson, S. Herzog, S. Jamin, Resource ReSerVation Protocol (RSVP) Version [6] A. Mankin, Ed., F. Baker, B. Braden, S. Bradner, M. O`Dell, A. Romanow, A. Weinrib, L. Zhang, Resource ReSerVation Protocol (RSVP) -- Version Applicability Statement Some Guidelines on Deployment, IETF RFC 228, September 997. [7] S. Blake, et al, An Architecture for Differentiated Services, RFC 2475, 998. [8] T. Li, Y. Rekhter, A Provider Architecture for Differentiated Services and Traffic Engineering (PASTE), IETF RFC 243, October 998. [9] Multi-Protocol Label Switching Architecture, RFC 33, Jan. 2. [] A Framework for Multi-Protocol Label Switching, Internet draft, draft-ietf-mpls-frameworko2.txt,997. [] D.Owduche,et al., A framework for Internet traffic engineering,draft-ietf-tewgframework-4.txt April 2. [2] D.O.Awduche et al., Requirements for Traffic Engineering over MPLS,RFC 272. [3] F. Le Faucheur, et al., Multi-Protocol Label Switching(MPLS)Support of Differentiated Services, RFC 327, 22. [4] E.Crawley et al. A framework for QOS-Based Routing in the Internet. RFC 2386. [5] Q. Ma and P. Steenkiste, Quality-of-Service Routing with Performance Guarantees., 4th International IFIP,Workshop on QOS. [6] S. Chen and K. Nahrstedt, On Finding Multi-Constrained Paths., IEEE ICC 98, June 998. [7]Bin Wang et al., A New Bandwidth Guaranteed Routing Algorithm for MPLS Traffic Engineering, IEEE, 22. P54/9