Designing a New Routing Simulator for DiffServ MPLS Networks Peng Zhang Zhansong Ma Raimo Kantola {pgzhang, zhansong, SCSC 2001 International Symposium on Performance Evaluation of Computer and Telecommunication Systems Orlando, Florida, July TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

Contents DiffServ MPLS network functions QoS Routing Performance and Cost Enhanced QoS Routing Simulator Link State Update Algorithms Simulation Study of LSU Algorithms TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

3 DiffServ Architecture TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY Customer Services Operation, management and provision QoS signaling DS Domain Ingress node Egress node Classifying, marking Interior node Forwarding PHB shaping

4 DS Traffic Engineering Functions TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY classifiermarkerqueuerdropper meter queuerdropper markerqueuerdropper Ingress node Interior node Egress node

5 Dropping and Queuing TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY EF queue AF1 queue AF2 queue AF3 queue BE queue AF4 queue WRED1 You can set: buffer size, serving rate, wred parameters etc...

Feasibility of QoS Routing QoS Routing – finds paths that are subject to QoS requirements while achieving efficiency in network resource utilization Benefits of QoS Routing – enables creation of virtual circuit-like services – improves user satisfaction – improves network utilization and increases network thruput Benefits do not come free – cost of deploying QoS routing protocol – cost of processing overhead – cost of storage overhead – added complexity of routers –... Whether or not the benefits are worth the cost ? TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

QoS Routing Performance QoS routing performance is usually described in terms of the utilization of network resources or the throughput achieved by network traffic. It is affected by such parameters: – routing algorithm complexity – link state updating algorithm: frequency of updating network state – network topology -- a certain routing algorithm may be better suitable for certain topologies, -- match between the traffic matrix and the topology – high level admission control: deciding when to accept a request TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

QoS Routing Cost Link state update cost – cost of generating and receiving LSAs – traffic cost Computation cost – cost of on-demand path computation – cost of path pre-computation Storage cost – storing more information than best effort routing – extra routing table if path pre-computation is used Other costs – signaling cost – software cost – operation cost – maintenance cost TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

Link State Update Algorithms Period based (PB) – timeout T – an update is sent when timeout T expires, T is restarted. Threshold based (TB) – threshold th, last advertised value bw i o and current value bw i c – updates when (|bw i o - bw i c | / bw i o )> th for bw i o  0, bw i c >0 for bw i o  0 Equal class based (ECB) – base class B, classes: (0,B),(B,2B),(2B,3B),…,etc – updating when available bandwidth changes across a class boundary Unequal class based (UCB) – base class B and factor f(f>1), classes: (0,B), (B,(f+1)B), ((f+1)B, (f 2 +f+1)B), ((f 2 +f+1)B,(f 3 +f 2 +f+1)B),…, etc – updating when available bandwidth changes across a class boundary TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

Multi-path QoS Routing TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY Motivation – The development of MPLS and DiffServ – The development of QoS routing (e.g., QOSPF) – The development of multi-path (QoS) routing in legacy networks – E.g., ATM networks Content – Find multiple disjoint paths between a pair of nodes that satisfy QoS requirements (e.g., bandwidth, delay) – Link disjoint paths or node disjoint paths – For example, two link disjoint paths have no common links – Fully disjoint paths and partly disjoint paths – Fully disjoint paths have no common links/nodes – Partly disjoint paths have common links/nodes but not all – On-line or off-line path computation – On-line computation uses simple algorithms – Off-line may use complex algorithms – Combination: Once link failures happen, the on-line computation is used first for fast restoration and the off-line computation is used later for path optimization

Multi-path QoS Routing (cont’d) TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY Benefits – Enhance network reliability and survivability – If one or more paths fail, other paths are available – If one or more paths fail, users’ traffic are not interrupted – Minimize network delay – Improve network resource utilization – Satisfy users’ traffic requirements Metrics – Simplicity – There lack simple and efficient multi-path routing algorithms – There are few multi-path QoS routing algorithms – Stability – Routing algorithms should efficiently work in various network environment, e.g., various network topologies, various traffic. – …

Multi-path QoS Routing (cont’d) TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY Considerations in QRS – Implement simple multi-path (QoS) routing algorithms – Inherit single path (QoS) routing algorithms of QRS – I.e., lowest cost (LC) and widest bandwidth (WB) – Based on link state schema (e.g., OSPF) Algorithms in QRS – Two-step-link-disjoint-lowest-cost – The first step: find a lowest cost path that satisfies the bandwidth requirement – The second step: prune the links used in the first path and then calculate the lowest cost path in the residue topology – Two-step-node-disjoint-lowest-cost – Two-step-link-disjoint-widest-bandwidth – Two-step-node-disjoint-widest-bandwidth Current work – Implement more multi-path QoS routing algorithms – Evaluate the performance of these algorithms and their impact on the network traffic – …

Enhanced QoS Routing Simulator This version supports DiffServ and MPLS TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY RM QOSPFERSVP Realtime Traffic Source/Sink NODE(Classifier, marker) LINK(dropper, queuer)

ISP Topology Link Bandwidth = 6Mb/s TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY Minimum cut is 5*6Mb/s = 30 Mb/s Shortest path algorithm is able to use 3 of 5 alternative links.

Simulations with UCB TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

Summary We have implemented a QoS Routing Simulator that supports IntServ, DiffServ and MPLS. It can be used to study dynamics, stability, cost and performance and scalability of QoS routing. We demonstrated the use of the Simulator for Unequal Class Based (UCB) link state updates. In UCB, the performance and cost are affected by two variables, by setting suitable values, this algorithm can produce a good combination of performance and cost. Most recently two multi-path QoS routing algorithms have been added to the Simulator. The Simulator is available for download on our www- site. TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

QoS Routing Protocol(QOSPF) TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY

QoS Routing Algorithms Lowest cost algorithm – finds a path p between source node and destination node such that the path bandwidth b(p)  B (required bandwidth) and the path has the least cost; – first eliminates all links that do not meet the bandwidth requirement; – then uses Dijkastra's algorithm to find the least cost path. Widest bandwidth algorithm – finds a path p between source node and destination node such that the path bandwidth b(p)  B (required bandwidth) and the path has the widest bandwidth; – first eliminates all links that do not meet the bandwidth requirement; – then uses the variation of Dijkastra's algorithm to find the the widest path. TEKNILLINEN KORKEAKOULU HELSINKI UNIVERSITY OF TECHNOLOGY