Skip to main content

Case Study 3 - BGP Rerouting of Prefixes


Due to the cable outages, some prefixes changed their routes for a significant time period. We analysed the effects of the SEA-ME-WE4 outage on the prefixes originated by AS17641 (Infotech, Bangladesh).

Routing States of a Prefix Originated by AS17641 - BGPlay screenshots

We looked at the routing dynamics of the prefix (originated by AS17641), using BGPlay. The following figures show some of the key routing changes the prefix underwent.

4:30 (UTC), 30 January 2008: Until this moment, not much has happened in terms of BGP messages for this prefix. The purple histogram at the left of the graph shows we are at the start of a period with high activity, triggered by one of the cable faults.


12:40 (UTC), 30 January 2008: Five hours after the event, BGP temporarily stabilises. Note that AS702 (UUNET Europe) is no longer seen in any route to AS17641. Also, only a few routes remain using the links through AS17494 (Bangladesh Telegraph & Telephone Board). All others switched to using AS17927 (WEBSATMEDIA PTE LTD, Satellite Over IP, Singapore) and AS17975 (APT Telecom Services Ltd., Hong Kong) for transit.


Cross Prefix Analysis - BGPath screenshots

We also used the BGPath tool to look at the routing dynamics of all prefixes originated by AS17641 as well as the dynamics of some inter-AS links seen in the AS paths ending at AS17641.

Path Evolution of All the Prefixes Originated by AS17641

AS17641 announces three prefixes.

These prefixes ( and, whose paths usually pass through AS17494, have been rerouted through AS17927 and AS17975. The prefix, which passes through AS24122 (BDCOM Online Limited, Bangladesh), completely lost its connectivity.


We analysed how the number of prefixes through AS links "related" to AS17641 (that is, adjacent to AS17641's upstream providers and announcing AS 17641's prefixes) changed.

30 January 2008: Fault

The BGPlay plots showed how AS17641 stopped using paths through 702-17494-17469 after the cable cut. At the same time, many other prefixes stopped being routed over these paths too. The following figures show how the number of prefixes announced by AS17641 and others through the AS links 702-17494 and 17494-17469 suddenly dropped. In particular, the AS link 702-17494, usually used by more than 200 prefixes, is not used by any of them for a significant time period. This is evidence of a major network event, something that affects the connectivity between AS702 and AS17494.



The BGPlay plots also showed that, as a consequence, AS17641's prefixes moved from 702-17494-17469 to 5511-17927-17469. Other prefixes also changed paths, preferring these two Autonomous Systems after the fault. As illustrated in the figures below, the number of prefixes announced by AS17641 and others through the AS links 5511-17927 and 17927-17469 increased.



Because almost all the prefixes affected by the event changed their paths without losing their connectivity, almost no change was observed on the AS link 17469-17641.


8 February 2008: Recovery

When the SEA-ME-WE4 cable was repaired and the original routing state recovered, we observed opposite changes in the number of prefixes on the AS links. As shown in the figure below, the link 702-17494 sees a sudden increase to it's old level of about 250 routed prefixes in the afternoon of 8 February 2008.



This analysis shows that from 30 January 2008 to 8 February 2008, the AS-link 702-17494 experienced some major event, and all the prefixes (whether originated by AS17494 or not) usually passing through this link were rerouted through different links. The timing of this event, 04:30 (UTC), aligns well with the recorded outages of the TTM box in Bahrain and other events in BGP. Because the FEA cable went down at 08:00, we conclude the AS link 702-17494 is set up over the SEA-ME-WE4 cable.