A Novel Monitor Placement Algorithm for Accurate Performance Monitoring in Optical Networks M. Angelou 1,2 , Y. Pointurier 3 , S. Azodolmolky 1,2 , D. Careglio 1 , S. Spadaro 1 and I. Tomkos 2 1 Universitat Polit` ecnica de Catalunya, C/ Jordi Girona, 1-3. 08034 Barcelona, Catalunya, Spain. 2 Athens Information Technology, 19.5km Markopoulou Ave., 19002 Peania, Athens, Greece. 3 Alcatel-Lucent, Bell Labs, Route de Villejust, 91620 Nozay, France. email: marianna@ac.upc.edu Abstract: We propose a novel monitor placement algorithm that reduces significantly the number of monitors required to accurately assess the QoT of all lightpaths established in an optical network, compared to other monitor placement methods. © 2010 Optical Society of America OCIS codes: (060.4250) Networks; (060.4510) Optical communications 1. Introduction Optical impairment and performance monitors are essential to ensure Quality of Service (QoS) in high capacity optical networks by providing failure management and Quality of Transmission (QoT) assessment capabilities [1]. A tech- nique presented in [2] allows to estimate end-to-end performance metrics of lightpaths (such as OSNR or Q factor) using only a small number of monitors, thereby enabling operators to reduce the associated cost of deployment and at the same time monitor accurately the network performance. In [3], we showed how monitoring devices can compen- sate the inaccuracy of analytical QoT computations in impairment aware routing decisions, but we did not specify a particular monitor deployment strategy. In this work we present an efficient heuristic to place a given number of im- pairment monitors in an optical network, where missing information is estimated using the linear estimator presented in [2]. Monitor placement is done at network design time, in order to facilitate performance monitoring at network operation time. We show on a 46-link network that our heuristic returns accurate performance estimates with only 15 monitors, down from 35 or more with other monitoring placement techniques. 2. Estimation Framework and Monitor Placement Heuristic Placing a monitor at the end of a lightpath provides a real-time measurement of an end-to-end performance metric of that lightpath (e.g. OSNR). To find the optimum monitor locations, the proposed algorithm utilizes the following idea: it is possible to use monitoring measurements collected from a subset of the network lightpaths to estimate accurately the corresponding end-to-end metric of the lightpaths that are not monitored. In order to estimate the QoT metrics using the partial monitoring data, we exploit an estimation framework known as “network kriging” (NK) [2]. NK is a linear estimator and applies to any “link-additive” performance metric. A performance metric is considered link-additive if it results from the sum of the metrics corresponding to the links traversed by the lightpath. The Chromatic Dispersion (CD; through the residual dispersion), the Polarization Mode Dispersion (PMD; through the average squared differential group delay), the nonlinear phase, and the inverse of OSNR are examples of physical impairments that are link-additive. Each of the mentioned metrics can be effectively measured by physical monitors. In fact the authors of [4] combined the aforementioned 4 parameters (i.e., CD, PMD, nonlinear phase, and 1/OSNR) to estimate the QoT of the lightpaths for establishment purposes. The monitor placement method presented herein is applicable to any link-additive metric; the results presented in the numerical section assume OSNR monitoring. In [2] the problem of monitor placement was tackled using an algebraic technique based on QR-decompositions of matrices (further denoted as the QR technique). Here we propose the “Pseudo-Eye” (PE) placement algorithm defined in Alg. 1. In PE it is assumed that each link is a possible monitor location and is fitted with either a real monitor or a “pseudo-monitor”. A pseudo-monitor uses the NK framework to estimate the corresponding performance metric based on the measurements from the real monitors located in other links. Given a network topology (graph G( V, E ), set of nodes V , set of links E ), a set of planned (routed) lightpaths L and a number m of (hardware) monitors that an operator is willing to deploy in the network, the algorithm computes the optimal (in terms of QoT estimation accuracy) locations for each of the m monitors. To achieve this, the algorithm first assumes that all network links are equipped with monitors, and then replaces the monitors that contribute least to the accuracy of the performance metric estimation by NK, with pseudo-monitors. JWA53.pdf OSA/OFC/NFOEC 2011 JWA53.pdf ©Optical Society of America