Re: Route Flap Damping: Harmful ?
Date: Wed, 18 Sep 2002 09:09:51 -0400 From: Zhuoqing Morley Mao <zmao@research.att.com>
Could you elaborate on the phenonmeon you observed? In particular, how were you able to determine it was caused by the BGP convergence effects?
Hi Morley, we (ACOnet/AS1853) do have our primary transit through Sprint/AS1239 and a (lowerpref) backup through Geant/AS20965_3549. We are receiving a full table from both (on two different routers). The path through Sprint for most of the US Internet is shorter than through Geant. When comparing the flap statistics on the two routers many of the listed prefixes are the same with about the same "duration", but the number of flaps seen through the longer path is much bigger (factor of 2-3 !) As I wrote in a previous mail, feel free to use our looking-glass http://www.aco.net/lookingglass/lg.html [flap statistics] to see the enormous amount of flaps counted by our Geant connected router. Unfortunately our Sprint connected router is not accessible through the looking-glass. But if you like I could produce snapshots and send them to you.
We hope that someone in Europe could also help out with the experiment, details of which are described at http://www.psg.com/~zmao
I think there was the idea to eventually enable RIS boxes as BGP beacons ? We do have a RIPE RIS box in Vienna at the VIX (RRC05).
We currently have two beacons in the U.S., some topology diversity would really help.
If RIS doesn't help we are considering setting up an ACOnet beacon for you, however this might not happen before begin of October.
Thanks!
You are welcome ! Kind regards CP --- ---------------------------------------------------------------------- --- --- Christian Panigl : Vienna University Computer Center - ACOnet --- --- VUCC - ACOnet - VIX : -------------------------------------------- --- --- Universitaetsstrasse 7 : Mail: Panigl@CC.UniVie.ac.at (CP8-RIPE) --- --- A-1010 Vienna / Austria : Tel: +43 1 4277-14032 (Fax: -9140) --- --- ---------------------------------------------------------------------- ---
At 07:16 PM 9/18/2002, Christian Panigl, ACOnet/VIX/UniVie wrote:
We currently have two beacons in the U.S., some topology diversity would really help.
If RIS doesn't help we are considering setting up an ACOnet beacon for you, however this might not happen before begin of October.
The RIS routing beacons are up now. Data is available. Excerpt from AmsIX->ACOnet: A 195.80.227.0/24 2002-10-01 00:00:43 193.203.0.1 1853 1239 6762 12654 W 195.80.227.0/24 2002-10-01 00:00:47 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:00:35 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:00:46 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:00:55 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:01:06 193.203.0.81 A 195.80.227.0/24 2002-10-01 02:01:15 193.203.0.1 1853 20965 1299 1239 6762 12654 W 195.80.227.0/24 2002-10-01 02:01:15 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:01:26 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:01:36 193.203.0.81 A 195.80.227.0/24 2002-10-01 02:01:42 193.203.0.1 1853 20965 1299 4200 3549 1239 6762 12654 W 195.80.227.0/24 2002-10-01 02:01:46 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:01:55 193.203.0.81 A 195.80.227.0/24 2002-10-01 02:02:12 193.203.0.1 1853 1239 701 6762 6762 12654 W 195.80.227.0/24 2002-10-01 02:02:26 193.203.0.81 W 195.80.227.0/24 2002-10-01 02:02:39 193.203.0.1 W 195.80.227.0/24 2002-10-01 02:02:55 193.203.0.81 Interesting: All those withdrawals from the peer that never announced the route. I knew that there are more withdrawals than announces. But more getting them without any announces is strange. Also I did expect the "fading out" to happen via monotonously increasing path lengths. But apparently somewhere there are places with higher hysteresis configured than others. Interesting. Would it be useful to vary the announce/withdrawal frequencies for dampening studies at this point or will people be busy analysing what we have? Daniel
On Tue, 1 Oct 2002, Daniel Karrenberg wrote:
Would it be useful to vary the announce/withdrawal frequencies for dampening studies at this point or will people be busy analysing what we have?
I agree with experimentation, but I think we should give people a few days to update their filters (not all our peers are already accepting 195.80.224.0/20 prefixes), then take data with this setup for a week or so to get a baseline _before_ changing anything. In order to make the data analyzable, we should also very carefully record when which parameters were in effect. Henk ------------------------------------------------------------------------------ Henk Uijterwaal Email: henk.uijterwaal@ripe.net RIPE Network Coordination Centre WWW: http://www.ripe.net/home/henk Singel 258 Phone: +31.20.5354414 1016 AB Amsterdam Fax: +31.20.5354445 The Netherlands Mobile: +31.6.55861746 ------------------------------------------------------------------------------ That problem that we weren't having yesterday, is it better? (Big ISP NOC)
participants (3)
-
Christian Panigl, ACOnet/VIX/UniVie -
Daniel Karrenberg -
Henk Uijterwaal (RIPE-NCC)