In message <199501251514.KAA03501@merit.edu>, yakov@watson.ibm.com writes:
Ref: Your note of Wed, 25 Jan 1995 05:57:53 -0500
Routing flaps considered harmful... A while ago Curtis proposed a scheme for holddowns that has a property that a holddown time associated with a route increases with the frequency of this route flapping -- so that frequently flapping routes getting progressively long timeouts.
Yakov.
Yakov, Thanks for remembering that. Do you also remember where I put the sources? :-) [need a bigger disk]. That work got put on a low priority and has never resurfaced to the top of the queue. I had written some code for rcp_routed and then put it on hold since gated was eminent. The plan at the time was to wait until Dennis got gated's BGP real solid and stable and then finish that code and hook it in. Cisco (Tony Li, maybe Paul) were not thrilled with the proposal because it added state (an empty 4 byte pointer for stable routes, a small struct for unstable ones). The BGP WG overall response was an enthusiastic yawn and encouragement to come back when some code was written, so it was never officially submitted. At the time that seemed reasonable since I was writing code and seemed likely to come back. I did find the source and put the files up for ftp in case anyone cares to look. It is at: ftp://ftp.ans.net/pub/papers/route-dampen.ps ftp://ftp.ans.net/pub/papers/route-dampen.txt Or use your favorite browser to visit the abstracts file first: ftp://ftp.ans.net/pub/papers/abstracts.html That was the proposal then. Do we want to resurect this? If so, why? Are we trying to convince someone to do an implementation? I wish I had the time to do one for gated BGP. I'm not sure Paul has the time and I'm not sure Cisco is convinced that they can afford an increase in state. This is a solution at the protocol implementation level, which is probably the best kind since it automatically kicks in when something flaps. Similar techniques could be applied to an IGP in originating LSAs. There is also the manual administrative solution of taking flakey links down and fixing problems when they occur. With the manual solutions, like aggregation, the provider needing the benefit is not the one that needs to address the problem. If we want to talk about the larger topic of keeping route flap under control, I'll come moderately prepared to talk about this if we want to bring it up. If Sean has some "how bad is it" slides, then maybe he can lead the discussion. I haven't had time to keep the external route flap reporting running all the time and don't have time to summarize it, but anyone that wants to do an analysis is welcome to the raw data. Curtis