Joint analysis of network incidents and intradomain routing changes

Abstract

This paper studies how intradomain routing instability relates to events in network trouble tickets for two networks: a VPN provider and the Internet2 backbone network. Our goal in performing this joint analysis of routing and trouble tickets is to better understand the likely underlying causes of intradomain routing instability. We develop a method to correlate trouble tickets with instability events and find that, although unplanned events last longer than scheduled maintenance, there is no single underlying cause for most instability, and that these causes differ across networks. In comparison to a similar study from Labovitz et al. from ten years ago, we find that, while certain causes of instability such as maintenance and circuit problems remain significant, power issues have become much less prevalent, and software-related problems have become more common.

GMPLS adaptive level of recovery

“…Even if their origins have somewhat evolved [6], many failures are predictable thanks to the monitoring of equipment health. Several online failure prediction techniques using failure tracking, symptom monitoring, errors reporting or even undetected errors auditing have already been tested on large scale systems [7].…”

“…More network oriented parameters such as the bit error rate or packet loss rate of a network interface need to be integrated but, above all, software malfunction symptoms need to be continuously monitored, such as extraordinary log entries, system errors, unreleased file locks, file descriptor leaking, data corruption, or the abnormal usage of CPU, memory and I/O that can indicate a memory leaking. At last, communications with external systems like the Intrusion Detection System (IDS) in case of hacker attacks, and particularly with the Network Management System (NMS), especially in case of maintenance activities, account for more than 30% of the troubleshooting events [6]. Moreover, the usage of machine learning techniques [7], using Bayesian networks, time series analysis, Support Vector Machines or Semi-Markov Processes are promoted to allow online failure prediction to evolve with time and to provide more accurate predictions.…”

Computing Minimal Update Sequences for Graceful Router-Wide Reconfigurations

“…Then, this function clones this current state of the mdag (l. 3) and repeatedly performs swallowing operations on a temporary copy in order to extract the new last constraint. At each iteration, the minimum ∆ value among the remaining nodes in the copy is extracted (l. 5) and the associated nodes are added to the next swallowing list (l. [6][7][8]. It iterates this way until no cycles appear.…”

“…Function UPMAX initializes the next swallowing list with each node in roots whose associated ∆ value is greater than m (l. [4][5][6]. Those nodes depitcs routers that have already satisfied loop constraints (if any).…”

“…Typically, reconfigurations at a router granularity are among the most frequent operations typically performed in IP networks [6]. For this reason, we focus on the problem of supporting graceful reconfigurations in commonly used LinkState (LS) Interior Gateway Protocols (IGPs) such as OSPF [7] or IS-IS [8].…”