Scalable Monitoring Heuristics for Improving Network Latency

Abstract

We consider a routing overlay in which the delay of a path can be obtained at some fixed cost by sending probe packets, and investigate the joint minimization of the probing cost and the routing delay. Assuming that link delays are modelled by Markov chains, this problem can be cast as a Markov Decision Process (MDP). Unfortunately, computing the exact solution of this MDP is prohibitively expensive due to the well-known "curse of dimensionality". In this work we propose two scalable approaches that are fast enough to provide efficient solutions on practical time scales. We analyze the complexity of both approaches, and evaluate their accuracy in small synthetic scenarios for which the optimal monitoring policy can be computed. Finally, the robustness and the scalability of the proposed solutions are analyzed using real delay data collected over the Internet.

A Markovian Approach for Improving End-to-End Data Rates in the Internet

“…We evaluate the average utility (see (2)) for four policies : optimal, myopic policy that optimizes the immediate cost only, a receding horizon policy (with a horizon of 3) and a decomposition based heuristic. For a description of the last two policies, we refer the reader to [6].…”

“…The problem formulation in this section is essentially the same as in [8,6] except that the quantity of interest is bandwidth instead of delay. Consider a single origin-destination pair and {1, 2, .…”

Cooperative Learning for Disaggregated Delay Modeling in Multidomain Networks

“…In the literature, some previous works have proposed different methods for monitoring the QoS. Because using active probes requires dealing with the introduced overhead, the authors in [11] targeted at reducing the monitoring load without compromising delay accuracy, whereas the authors in [12] proposed to use active monitoring during commissioning testing to avoid such overhead. In the later work, the authors injected packet trains reproducing the real traffic that the service will support, and measurement data were used afterwards to produce a specific delay model for the packet connection that could be used during connection operation time.…”

“…Specifically, a decision score s is defined based on key statistical quartiles [30] of the average bias of every path in time. Equation (11) formally describes the computation of the quartiles 25%, 75%, and 100% % of the bias of all paths. The obtained results are then used to compute s in Eq.…”