What would be your recommendation to speed up dual-homed Internet BGP convergence and minimize the effect of ISP flapping? The usual stuff is in place, such as AS path prepend and RFC1998 ISP community. The convergence time takes about as little as 5 min to as much as 10 min.
It’s about 5 minutes from losing access to access being restored, during which we’re waiting for Internet routing convergence when one of the BGP peering is down. We enabled ping on the ingress Citrix rule – that’s how we measure whether the routing has been restored and ping starts to respond.
Our ISPs (Zayo/Cogent) all have looking-glass sites. I didn’t think about using them because I thought the issue would depend on the intermediary ISP. I have been using route-views.routeviews.org, but it shows prefixes to one ISP only.
We’ve had an issue with our ISP and its upstream peer in that the upstream peer always prefers (LP). What the intermediary ISP does is out of our hands. We have since moved away to another ISP, but I feel that BGP alone is insufficient to have a better than good high availability solution. I would personally accept and live with a 5-minute outage, but the client wouldn’t. I was wondering whether I missed any latest config knob or ISP feature that I’m not aware of.
Along with the redundant BGP internet routers, I’ve always wondered if there was a template of sort for internet facing routers from a enterprise perspective that would cover security and filters that should always be applied, like to keep your AS from becoming a transit AS and just plain old best practice.
How about the various design suggestions for using bgp in order to use multiple internet providers for a single site? or multiple sites?
How about routers receiving full routing tables from both ISP vs one router receiving a default route and the other one receiving a full routing table? Outbound traffic engineering strategies etc.
Typical in IXP but not as well documented as iBGP route reflectors.
If I get it right, ideally, they reflect the eBGP routes but not adding their own ASN (transparently); they calculate the best route on behalf of the participants (“Maintain a separated Routing Information Base (Loc-RIB) for each peer configured as RS-client”) and they can also filter bgp routes in/out but like they were the participants.