North American Network Operators Group|
Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical
Re: how are backups implemented?
let me repeat my question, this time more clearly. we, at uw, are analyzing bgp tables for possible errors (misconfigurations). one of the strange things (question 3 below) we are observing is the following. a prefix 10.10.0.0/16 (for instance) is announced by AS X. sometimes, some of its more-specifics (like 10.10.1.0/24, 10.10.56.0/24 ....) would appear for a short time (for example, 4 hours) and then disappear again. furthermore, these more-specifics would have an origin AS Y (Y != X). i am curious if this behavior can be caused by some sort of backup arrangements i don't understand, or some router/administrator mess-up. clues? thanks, -- ratul On Thu, 20 Sep 2001, Ratul Mahajan wrote: > > > [posting this message after having looked for answers elsewhere including > the archives, but found no satisfactory answers] > > i wanted to ask the operations community about how backups are typically > implemented. i am more interested in backup implementations, in which a > failure would expose a different origin AS (this would exclude prepending > based backups). > > 1. when a network is multihomed, and one of the links fails, would you > expect a smooth transition (as seen in the bgp tables of a remote AS) from > one origin AS to another (modulo convergence effects)? > > 2. can a failure (anywhere in the network) ever expose another origin AS > for some AS's while it stays the same for some? i guess it can, when the > network is being persistently announced from both origins, and under > normal scenario one origin could be hidden from some AS's. would this also > hold for a routing table as rich as routeviews? > > 3. can a failure ever cause more-specifics with a different (from the > origin of the less-specific) origin AS to appear (again, as seen from a > remote AS)? this might depend on how backups are implemented - so what i > am asking is, is this a common/possible case? > > > thanks, > -- ratul > >