Traffic Engineering TE 1 Network Congestion Causes of

Traffic Engineering (TE) 1

Network Congestion • Causes of congestion – Lack of network resources – Uneven distribution of traffic caused by current dynamic routing protocols • Consequences of congestion – High loss rate – Low throughput – Long end-to-end delay • Intserv and Diffserv provide differentiated degradation of performance for different traffic when the network is congested 2

Traffic Engineering • Traffic Engineering (TE) is the process of distributing traffic flows through the network to achieve load balancing • TE leads to: – Reduced congestion – Improved bandwidth utilization 3

TE Approaches • Preplanned: – OSPF + smart link weight setting – MPLS + optimal general routing • On demand – MPLS + Constraint-Based Routing 4

OSPF Routing • Each link has a static link weight configured by the network operator. – Examples: unit weight, weight proportional to physical distance of link, weight inversely proportional to link capacity • Packets routed over the shortest path to destinations – When multiple shortest paths exist to a destination, traffic is split evenly among the paths • Drawback: may cause uneven distribution of traffic 5

OSPF Routing • Routing depends on the choice of link weights Can control the distribution of traffic in the network by tuning the link weights. 6

Weight Tuning in OSPF • All links have same capacity, nodes q, r, s, w each has one unit of traffic to send to node t. • Objective: minimize the maximum link load. 7

Optimization of OSPF Link Weights • Given a network topology and a traffic matrix, find an optimal setting of the link weights so that a certain objective is achieved • Example objectives – Minimize the maximum link utilization (link utilization = link load/link capacity) – Minimize total cost of all links where the cost of a link is a function of link utilization 8

Optimization of OSPF Link Weights • Local search heuristic [Fortz and Thorup 2000] – Finding: For real networks, a good setting of the link weights can make OSPF perform almost as well as optimal general routing • General routing: traffic flow between nodes s and d can be split arbitrarily over the paths between s and d – Achievable with MPLS 9

Traffic Trunk • A traffic trunk is an aggregation of traffic flows belonging to the same class that are placed inside a LSP • Attributes of a traffic trunk – Qo. S requirements – Policy: include/exclude certain links 10

Constraint-Based Routing (CBR) • Given a traffic trunk, compute a path for it subject to multiple constraints – Qo. S constraints – Resource availability constraints – Policy constraints • Goals of CBR: – Meet Qo. S requirements of the traffic trunk – Increase the utilization of the network • MPLS can setup LSPs along paths determined by CBR 11

Routing Metrics • Let d(i, j) be a metric for link (i, j). For any path P = (i, j, k, …, l, m), metric d is: additive if d(P) = d(i, j) + d(j, k) + … + d(l, m) – delay, jitter, hop-count multiplicative if d(P) = d(i, j) * d(j, k) * … * d(l, m) – reliability (i. e. , 1 -loss rate) concave if d(P) = min{d(i, j), d(j, k), …, d(l, m)} – bandwidth 12

Complexity of CBR • Computing a route subject to constraints of two or more additive and/or multiplicative metrics is NP-complete. • The computationally feasible combinations of metrics are bandwidth and one of the other metrics. 13

Path Computation • Bandwidth and hop-count constraints are commonly used in path computation – Many real-time applications will require a certain amount of bandwidth. – The amount of resources consumed by a flow is proportional to the number of hops it traverses • Path Computation algorithm: Step 1. Prune links if: – insufficient bandwidth – violate policy constraints Step 2. Compute shortest path 14

Information Requirement of CBR • Information needed by CBR: – Network topology – Available bandwidth on links • Routers need to distribute new link state information, i. e. , link available bandwidth – Extend the link state advertisements of routing protocols (OSPF, IS-IS) 15

Information Distribution • Flooding link state advertisements whenever a link’s available bandwidth changes is too expensive • A tradeoff must be made between the accuracy of link available bandwidth information and the frequency of flooding of link state advertisements. 16

Information Distribution • Periodic scheme – Periodically, a node checks if the current link status is the same as the one lastly broadcasted – If different, floods updated links status • Threshold scheme: flood LSA on significant changes of available bandwidth (e. g. , more than 50% or more than 10 Mbps) • On topology changes: link addition/removal, link down/up 17

Information Distribution • LSP setup may fail due to inaccurate link information • When a node refuses to setup an LSP due to insufficient link bandwidth, it broadcasts an update of its available bandwidth 18

Tradeoff Between Resource Conservation and Load Balancing • Widest-shortest path routing: choose a path with min hop-count; if more than one such path, choose the one with the largest available bandwidth – Emphasize preserving network resources • Shortest-widest path routing: choose a path with largest available bandwidth; if more than one such path, choose the one with the min hop-count – Emphasizes load balancing 19