California fault lines

Abstract

Of the major factors affecting end-to-end service availability, network component failure is perhaps the least well understood. How often do failures occur, how long do they last, what are their causes, and how do they impact customers? Traditionally, answering questions such as these has required dedicated (and often expensive) instrumentation broadly deployed across a network.We propose an alternative approach: opportunistically mining "low-quality" data sources that are already available in modern network environments. We describe a methodology for recreating a succinct history of failure events in an IP network using a combination of structured data (router configurations and syslogs) and semi-structured data (email logs). Using this technique we analyze over five years of failure events in a large regional network consisting of over 200 routers; to our knowledge, this is the largest study of its kind.

Siamese Graph Attention Networks for robust visual object tracking

“…Fig. 1 shows the distribution of failure impact (i.e., maximum link utilization (MLU) increase) under OSPF 2 and optimal (MCF) routing scheme for three real-world large scale network topologies with more than 100 nodes. It turns out that only 0.19%, 0.03%, and 3.43% failure scenarios on Ion, Interoute, and DialtelecomCz, respectively, under the optimal routing scheme, cause significant impact (i.e., more than 80% of worst-case failure impact) to the network availability.…”

“…As the scale and complexity of modern networks continue to increase rapidly, the occurrence of failures has become a common and frequent event in both wide area networks (WANs) [1], [2], [3], [4], [5], [6] and data center networks (DCNs) [7]. Recently, increasing research efforts have considered tackling the problem from different aspects.…”

Advancing Network Monitoring and Operation with In-band Network Telemetry and Data Plane Programmability

“…In view of that, in Algorithm 7, whenever a valid notification is received, the procedure will also verify whether the notification was received within a reasonably short amount of time of the previously detected local failure (Lines 7-8). According to measurement studies (GOVINDAN et al, 2016;TURNER et al, 2012;GILL;JAIN;NAGAPPAN, 2011;TURNER et al, 2010;MARKOPOULOU et al, 2008), the majority of times, when two links fail simultaneously, they belong to the same shared-risk group. With that observation in mind, whenever a switch is aware of two single-link failures happening in a short period in time, it will transition to a tactic that handles the shared-risk group to which the two links belong by looking up table SF-TACTICS (L.9-11).…”

“…Consequently, computing alternative forwarding entries for all possible failure scenarios would both take an impractically long time and require prohibitive amounts of memory. Despite that, network measurement studies (GOVINDAN et al, 2016;TURNER et al, 2012;GILL;JAIN;NAGAPPAN, 2011;TURNER et al, 2010;MARKOPOULOU et al, 2008) show that although failures happen frequently (a few minutes apart), it is very unusual for two elements (e.g., links, switches, optical-fiber cable) to fail at the "same time", unless they belong to the same shared-risk group. For example, two links connected to the same switch are perceived as failed whenever the switch itself fails.…”

The Hazard Value: A Quantitative Network Connectivity Measure Accounting for Failures

“…There already exists interesting literature on the empirical characteristics of failures, e.g., in datacenters [12], [31], statewide networks [29], or IP backbones [16]. This literature is highly valuable for the comparison of existing networks, but does not directly solve the problem of comparing network designs that are not yet implemented.…”