RIPE has issued ripe-378
It suggests that 2 major problems arise from implementation of route-flap damping:
1) Impeded convergence
Basically a ripple effect, caused by the flap penalty with a change in BGP attribute (in this case, a new path to the same AS). The prefix probably has not been withdrawn from the routing table at all, but route-flap damping can be triggered at the 2nd remove - i.e. at the neighbour of the router/AS making the routing change. Not mentioned, but tunable on some implementations, is that attribute changes can incur a lower penalty than straight prefix withdrawals. This could help alleviate local problems, but will start causing inconsistent (and difficult to diagnose) path selection.
2) Transit of updates
Related to the consistency of implementation of RFD: timing of routing updates. Because of differences in the Minimum Route Advertisement Interval (MRAI) Timer (the amount of time a router waits before passing on a route update), as well as differences in paths, multiple updates for the same 'flapping' AS can arrive at a router many hops away and cause multiple updates as the router evaluates the new routes available. This can easily trigger flap damping. There have been actual measurements where this resulted in a single prefix withdrawal producing 41 BGP events a few hops away!
Result: ripe-378 recommends against RFD, citing higher router CPU power as able to cope with flap events without damping.
Yay Stupid Network!
No comments:
Post a Comment