Re: Withdrawls and announcements attempt 2
Curtis Villamizar <curtis@ans.net> Fri, 21 June 1996 22:05 UTC
Received: from ietf.cnri.reston.va.us by IETF.CNRI.Reston.VA.US id aa29113;
21 Jun 96 18:05 EDT
Received: from CNRI.Reston.VA.US by IETF.CNRI.Reston.VA.US id aa29109;
21 Jun 96 18:05 EDT
Received: from p-o.ans.net by CNRI.Reston.VA.US id aa16926; 21 Jun 96 18:05 EDT
Received: (from majordom@localhost) by p-o.ans.net (8.7.5/8.7.3) id VAA27326
for bgp-outgoing; Fri, 21 Jun 1996 21:53:01 GMT
X-Authentication-Warning: p-o.ans.net: majordom set sender to bgp-owner using
-f
Message-Id: <199606212150.RAA04203@brookfield.ans.net>
To: Yakov Rekhter <yakov@cisco.com>
Cc: curtis@ans.net, bgp@ans.net
Subject: Re: Withdrawls and announcements attempt 2
In-Reply-To: Your message of "Fri, 21 Jun 1996 10:34:12 PDT."
<199606211737.KAA29189@hubbub.cisco.com>
Date: Fri, 21 Jun 1996 17:50:31 -0400
Sender: ietf-archive-request@IETF.CNRI.Reston.VA.US
From: Curtis Villamizar <curtis@ans.net>
X-Orig-Sender: bgp-owner@ans.net
Precedence: bulk
Reply-To: bgp@ans.net
Yakov, Moved to bgp list. > Please note that propagating withdrawals to all neighbors is > *not* the problem we are trying to solve now, as such propagation > accounts for only a small fraction of the total withdraws (see attached). > > In fact, we've yet to see any empirical evidences that propagating > withdrawals to all neighbors is a problem. At MaeEast in many cases 30 to 50 peers hear a withdrawl and then withdraw the same route thenselves. You would expect to see N times as many withdraws as was needed for a given withdrawl depending on who withdrew it and how many Ciscos they peer with. > Now could we return (perhaps on the BGP list) to the discussion > about the *real* issue we need to solve - ~5*10^6 daily withdraws. > > Yakov. If you look at the stats and add up the withdraws and announces and take the ratio you have a ratio of about 12. Someone with a Cisco router that hears a lot of withdraws will have a very high ratio. Someone that flaps a lot will just cause others to have a very high ratio. The ratio of totals should be in line with the average number of Cisco peers. This should be sufficient emperical evidence to consider fixing the problem. The ratio of 12:1 out of does not seem out of line fora weighted average number of the number of Ciscos that a given flapper flaps to. I think some of the smaller providers do have less peers than the major providers, so this does seem reasonable to me. Curtis Peer Packets Ann With Ann+With ANS (690) 2572 8226 5732 13958 Advantis (2685) 83 76 59 135 Alternet (701) 16370 85551 145672 231223 BBNplanet (AS1) 4223 5643 22486 28129 BayNetworks (6205) 241 306 321 627 CAIS (3491) 3043 1285 197688 198973 CWInet (4445) 2875 67 267503 267570 CWix (5683) 2317 527 31636 32163 CWix (5683) 2970 486 141007 141493 Compuserve (4183) 18 3 19 22 DIGEX (2548) 1442 253 9323 9576 DRAnet (1746) 2 2 0 2 DialNet (5552) 85 35 113 148 ESnet1 (293) 237 130 682 812 ESnet2 (293) 243 282 509 791 EUNet (286) 1931 7167 4577 11744 Erols (6079) 869 100 11438 11538 Exodus (3967) 1443 16884 16860 33744 HLCnet (4565) 2685 11 112495 112506 INAP.net (5646) 4017 1276 115737 117013 INSnet (5378) 588 5 7812 7817 IOSnet (5000) 23 20 7 27 IconNet (3951) 12 9 15 24 Interpath (3407) 2895 174 266051 266225 MAI (6082) 8 29 0 29 MCI (3561) 8687 24039 198729 222768 Net99 (3830) 2858 0 217279 217279 Netcom (2551) 7040 23747 213759 237506 Netrail (4006) 29 54 12 66 PIPEX (1849) 14858 2890 366578 369468 PrimeNet (3549) 15 47 29 76 SPRINT (1239) 10561 34816 94378 129194 USCyber (4136) 2921 185 157816 158001 VoiceNet (3932) 60 57 373 430 Totals 214382 2606695 ratio: 12.2
- Re: Withdrawls and announcements attempt 2 Curtis Villamizar
- Re: Withdrawls and announcements attempt 2 Yakov Rekhter