Modern communication networks are able to respond to randomly uctuating demands and failures by allowing bu ers to ll, by rerouting tra c and by reallocating resources. They are able to do this so well that, in many respects, largescale networks appear as coherent, almost intelligent, organisms. The design and control of such networks present challenges of a mathematical, engineering and economic nature. In this lecture I describe some of the models that have proved useful in the analysis of stability, statistical sharing and pricing, in systems ranging from the telephone networks of today to the information superhighways of tomorrow.

IP forwarding anomalies, triggered by equipment failures, implementation bugs, or configuration errors, can significantly disrupt and degrade network service. Robust and reliable detection of such anomalies is essential to rapid problem diagnosis, problem mitigation, and repair. We propose a simple, robust method that integrates routing and traffic data streams to reliably detect forwarding anomalies, and report on the evaluation of the method in a tier-1 ISP backbone. First, we transform each data stream separately, to produce informative alarm indicators. A forwarding anomaly is then signalled only if the indicators for both streams indicate anomalous behavior concurrently. The overall method is scalable, automated and self-training. We find this technique effectively identifies forwarding anomalies, while avoiding the high false alarms rate that would otherwise result if either stream were used unilaterally.

We show that probability distributions and moments of performance measures in many polling models can be effectively computed by numerically inverting transforms (generating functions and Laplace transforms). We develop new efficient iterative algorithms for computing the transform values and then use our recently developed variant of the Fourier-series method to perform the inversion. We also show how to use this approach to compute moments and asymptotic parameters of the distributions. We compute a two-term asymptotic expansion of the tail probabilities, which turns out to be remarkably accurate for small tail probabilities. The tail probabilities are especially helpful in understanding the performance of different polling disciplines. For instance, it is known that the exhaustive discipline produces smaller mean steady-state waiting times than the gated discipline, but we show that the reverse tends to be true for small tail probabilities. The algorithms apply to describe the transient behavior of stationary or nonstationary models as well as the steady-state behavior of stationary models. We demonstrate effectiveness by analyzing the computational complexity and by doing several numerical examples for the gated and exhaustive service disciplines, with both zero and non-zero switchover times. We also show that our approach applies to other polling models. Our main focus is on computing exact tail probabilities and asymptotic approximations to them, which seems not to have been done before. However, even for mean waiting times, our algorithm is faster than previous algorithms for large models. The computational complexity of our algorithm is O(N α) for computing performance measures at one queue and O(N 1 + α) for computing performance measures at all queues, where N is the number of queues and α is typically between 0.6 and 0.8.

this article, we will use the term network reliability in a broad sense and cover several subtopics. We will start with network availability and performability, and then discuss survivable network design, followed by fault detection, isolation, and restoration as well as preplanning. We will conclude with a short discussion on recent issues and literature.

TINA is an open, object oriented, distributed telecom architecture, with many concepts taken directly from the latest computer research. In TINA, instances of the same object type can be placed on different physical nodes. Therefore, the network performance can be improved by introducing load balancing algorithms. These algorithms should distribute the traffic between the object instances in such way that the overall throughput and setup time are improved. We discuss and examine a number of simple distributed load balancing algorithms, that do not require any extra load information exchange between the nodes. The results show that it is difficult to find an algorithm that behave well for all traffic situations. The main problem is that the algorithms have not enough information about the load situation on the different nodes, since no load information is exchanged between the nodes. This problem can be solved by adding the feasibility of load status information to the TINA protocols. 1.

Abstract: Poison message failure propagation is a mechanism that has been responsible for large scale failures in both telecommunications and IP networks. We propose a combination of passive diagnosis and active diagnosis to deal with the poison message problem. This paper focuses on the active diagnosis problem in which filters are dynamically configured to block suspect protocols or message types. While message filtering can potentially interrupt failure propagation and aid in identifying the poison message type, it can negatively impact the control and management of the network. We formulate this tradeoff as a Markov decision process (MDP) whose solution is the optimal policy for determining which message types to filter at each time step. The size of the state space makes it impractical to use traditional techniques to solve the MDP. Consequently we use a combination of reinforcement learning and feature-based Q-factor approximation to obtain a suboptimal policy. Simulation based experiments indicate that this policy performs significantly better than a heuristic policy that is well known in fault management applications. 1.

Poison message failure propagation is a mechanism that has been responsible for large scale failures in both telecommunications and IP networks: Some or all of the network elements have a software or protocol ‘bug ’ that is activated on receipt of a certain network control/management message (the poison message). This activated ‘bug ’ will cause the node to fail with some probability. If the network control or management is such that this message is persistently passed among the network nodes, and if the node failure probability is sufficiently high, large-scale instability can result. Identifying the responsible message type can permit filters to be configured to block poison message propagation, thereby preventing instability. Since message types have distinctive modes of propagation, the node failure pattern can provide valuable information to help identify the culprit message type. Through extensive simulations, we show that artificial neural networks are effective in isolating the responsible message type. Poison message, neural network, node failure pattern, fault management 1.

In this paper, we present a framework of fault management for a particular type of failure propagation that we refer to as “poison message failure propagation”: Some or all of the network elements have a software or protocol ‘bug’ which is activated on receipt of a certain network control/management message (the poison message). This activated ‘bug ’ will cause the node to fail with some probability. If the network control or management is such that this message is persistently passed among the network nodes, and if the node failure probability is sufficiently high, large-scale instability can result. In order to mitigate this problem, we propose a combination of passive diagnosis and active diagnosis. Passive diagnosis includes protocol analysis of messages received and sent by failed nodes, correlation of messages among multiple failed nodes and analysis of the pattern of failure propagation. This is combined with active diagnosis in which filters are dynamically configured to block suspect protocols or message types. OPNET simulations show the effectiveness of passive diagnosis. Message filtering is formulated as a sequential decision problem, and a heuristic policy is proposed for this problem. 1.