MATE: MPLS Adaptive Traffic Engineering Anwar Elwalid, et. al. IEEE INFOCOM 2001

Contents Introduction MATE Functions and Algorithms MATE Implementation Techniques Simulation Results Conclusions

Introduction (1/3) Traffic engineering (TE) v.s. QoS routing –TE aims at maximizing operational network efficiency while meeting certain constraints –QoS routing meet certain QoS constraints for a given source-destination traffic flow Two categories of TE implementation –Extend current shortest path algorithm based routing protocol, e.g. OSPF-TE –MPLS based TE, e.g. RSVP-TE, CR-LDP

Introduction (2/3) Limitations of extending SPF-based routing –Load sharing can not accomplished among paths of different costs –Traffic/policy constraint are not taken into account –Modifications of link metrics to re-adjust traffic mapping tend to have network-wide effects –Traffic demands must be predicable and known a priori The combination of MPLS technology and its TE capabilities are expected to overcome the above limitations.

Introduction (3/3) MPLS TE mechanisms may be –Time-dependent mechanisms use historical information based on seasonal variations in traffic to pre-program LSP layout and traffic assignment do not attempt to adapt to unpredictable traffic variations or changing network conditions –State-dependent mechanisms Deal with adaptive traffic assignment to the established LSPs according to the current state of the network –The focus of this paper is on load balancing short-term traffic fluctuations among multiple LSPs between an ingress node and an egress node

MATE Functions & Algorithms (1/4) MATE functions in an ingress node

MATE Functions & Algorithms (2/4) Filtering and Distribution function –Facilitate traffic shifting among LSPs in a way that reduces the possibilities of having packets out of order Traffic Engineering function –Decides on when and how to shift traffic among LSPs –Consists of two phases: monitoring phase and engineering phase Measurement and Analysis function –Obtains one-way LSP statistics such as packet delay and packet loss, done by having ingress node transmit probe packet periodically to the egress node which returns them back to ingress node

MATE Functions & Algorithms (3/4) Model –L: a set of unidirectional links, shared by –S: a set of ingress-egress(IE) node pairs, each pair s has –P s : a set of LSPs –An IE pair s has total input traffic rate r s and route x sp amount of it on LSP p such that  p  Ps x sp = r s, for all s –x l : flow rate on link l  L, –C l (x l ): cost function of link flow x l –Objective:

MATE Functions & Algorithms (4/4) Asynchronous algorithm –Gradient projection algorithm: iteratively adjusted in opposite direction of the gradient and projected onto the feasible space. Each iteration takes the form x(t+1) = [x(t) -  C(t)] +,where  >0 is a stepsize, should be chosen sufficiently small  C(t) is a vector whose (s,p)th element is  C/  x sp [z]+ is the projection of a vector z onto feasible space –The algorithm terminates when there is no appreciable change, i.e.,||x(t+1)-x(t)|| < 

MATE Implementation Techniques Traffic filtering and distribution –Distribute traffic on a per-packet basis without filtering –Filter traffic on a per-flow basis and distribute the flows to the bins such that the loads are similar –Filter the incoming packets by using a hash function Traffic measurement and analysis –Packet delay and packet loss probability are metrics that can be estimated by a group of probe packets –Bootstrap technique is used to dynamically select the required number of probe packet to send

Experimental Methodology Two network topologies Two types of traffic: engineered traffic and cross traffic Two traffic models: –Short-term dependencies: Poisson –Large degree of dependencies: DAR Implementation of the algorithm –Random delay introduced before moving from the monitoring phase to the traffic engineering phase –Coordination among ingress nodes Network topology 1 Network topology 2

Poisson traffic for network topology 1DAR traffic for network topology 1

 With cross traffic and  engineered Poisson traffic  for network topology 2

Poisson traffic with coordinationDAR traffic with coordination

Conclusions MATE algorithms are proposed –To apply adaptive TE to utilize network resource more efficiently and minimize congestion –Using minimal assumptions through a combination of techniques such as bootstrap probe packets –With stability and optimality proved by analytical models –To effectively remove traffic imbalances among multiple LSPs from simulation results