Lecture 15 IGP and MPLS D Moltchanov TUT

Lecture 15. IGP and MPLS D. Moltchanov, TUT, Spring 2015

Outline IP intra-domain traffic engineering MPLS tunneling optimization

IP intra-domain traffic engineering

IP Intra-domain TE Routing in the Internet Intra-domain (interior gateway protocols, IGP) Inter-domain (exterior gateway protocols, EGP) Intra-domain routing Performed by ISP controlling its own network Efficiently for cross-traffic and own traffic IGP Routing protocols Open shortest path first (OSPF, IETF) Intermediate system to intermediate system (IS-IS, ITU-T) Link-state protocols with similar functionality Link-state: use link-weight metrics to determine shortest path Provides common view for all the routers in the networks

IP Intra-domain TE An example of general problem How to minimize the delay for packets traversing the network What would be the appropriate way of doing so? Delays happens due to link congestion Suppose happens on the shortest path for a demand Link weight is too low on a link? Shortest path should not use this link How? Increase the link weight! OSPF/IS-IS will use other link instead of this one General problem Determine the links weights such that SPs are determined in a way minimizing the delay! Which delay? Average packet delay, not individual one On all links individually? Average over all links?

IP Intra-domain TE Input data Network itself: graph having V nodes, E links Demands between routers: Link capacities: Design objective Metric related to delay but which one? Minimize the maximum link utilization! What are we doing with this? Minimizing delay On all SPs simultaneously! The problem now Minimize the maximum link utilization, given that we have demands and link capacities and the routing happens according to OSPF/IS-IS

IP Intra-domain TE Formulating the problem mathematically Let the link weights system be It induces all the flows in the network OSPF/IS-IS: SP routing with “equal-cost multi-path” rule (ECMP) Let be the set of all path for demand d Not necessarily SP as we do not know yet Can be done using K-shortest path algorithm Let flow of path p for demand d induced by We have demand constraints Isn’t is similar to what we got solving NDP problem? Yes, but flows depend on the link metric system !

IP Intra-domain TE Getting capacity constraints is next Let be the link path indicator if route p for demand d uses link e, 0 otherwise Link load of link e is given by And it should be less than given capacity of link e What we got so far? Demand constraints Capacity constraints

IP Intra-domain TE What is our objective? Minimize maximum utilization of the link Utilization of a link e is The maximum utilization is then The whole problem: minimize subject to: for integers over

IP Intra-domain TE Can also be stated as Minimize Subject to over and for integers and continuous If optimal then no links are congested Practice: task is solved in off-line and then Why? No need to implement on-line are disseminated Complex to solve! There might be no solutions!

IP Intra-domain TE What is about relaxed model? What if no constraints are imposed by ? This means when allocations are free of dependence on Does it has any significance for SP problems like this? Why we care? That would be LP problem Results for relaxed model In general far from optimal Link metric system induces strict constraints on

MPLS tunneling optimization

MPLS tunneling optimization Why MPLS? MPLS: multi-protocol label switching No hop-by-hop routing anymore Virtual paths (AKA tunnels) instead are used An individual virtual path for a traffic class No more IP routing in domains A virtual class is uniquely identified by labels Why MPLS is so good? We can route as we want Not only along shortest paths Can work with Diff. Serv Can provide Qo. S (together with Diff. Serv) and network control… Huge popularity in US (see Verizon, Comcast, etc. )

MPLS tunneling optimization MPLS tunnels Also called label switched path (LSR) Established using label distribution protocol (LDP) Without MPLS TE: just mapping from OSPF With MPLS TE: whatever we want MPLS TE features: RVSP-TE or LDP-TE We need to know how and where to set-up tunnels Problems? Number of tunnels: should be kept at minimum How and where to establish tunnels such that load is balanced Task statement How to carry different traffic classes in MPLS network though the creation of tunnels in such a way that the number of tunnels on each MPLS switch is minimized while the load is balanced

MPLS tunneling optimization Formalization of the problem d denotes a demand (depends on traffic class and pair of nodes) is the demand volume a demand can be carried over different tunnels (splitting allowed!) Do the following let be the number of different paths allowed for d let be the fraction of demand d routed over p We get demand constraints Notice different use of flow variable We may get plenty of flows with small allocation… Recall we want to keep the number of tunnels at minimum!

MPLS tunneling optimization What to do with “small” allocations? Introduce a minimum constraint (lower bound) ? letting : 1 if higher than and 0 otherwise How to use and to enforce no “small” flows? If tunnel is chosen: greater than If not (less than ): force to 0 Capacity constraint: Number of tunnels on a link

MPLS tunneling optimization Objective function Minimize number of tunnels to carry traffic Let r be the max number of tunnels over all links The problem Minimize Subject to continuous non-negative, binary, integer Continuous and discrete variables? Mixed integer problem! Complex… Can be extended to limit number of tunnels on a single link!