nanog mailing list archives

Re: Filtering levels (was RE: multi-homing without the BGP (was RE: Packet Loss))


From: "Daniel L. Golding" <dan () netrail net>
Date: Sun, 17 Dec 2000 11:46:45 -0500 (EST)


Travis,

I think the way to fix this is to use non-summary aggregates in
combination with reasonable prefix limit filters - nothing smaller than
/24. This is a fair combination of cutting down on extraneous route
advertisements, and preserving customer route informations. I certainly
agree that advertising /30s to peers is unacceptable. How does it go? "Be
liberal in what you accept, be conservative in what you advertise".

- Daniel Golding

On Sun, 17 Dec 2000, Travis Pugh wrote:


Hi Daniel. 

You're entirely correct, in the case of a multihomed customer who
needs routing information propagated.  However, I differentiate between
wholesale deaggregation and an ISP that takes care of multihomed
customers' needs.

To me, there's a believability factor in people claiming that they do
large-scale deaggs for their customers.  Part of that believability is
doing due diligence on what they announce, and part of it is doing due
diligence on what their customers announce.  It doesn't appear that any of
that has been done here, which makes me skeptical.  A bunch of /25 and
greater adverts from 6347, and a bunch of /30s from customer ASNs, won't
convince me that anyone needs to relax their filters any time soon.

Some snippets from nitrous:

*>i64.240.169.224/27 209.44.160.105  4294967294    100      0 6347 i
*>i64.241.182.128/25 209.83.160.22   4294967294    100      0 6347 i
*>i64.242.80.0/27   209.44.160.105  4294967294    100      0 6347 i
*>i209.44.23.160/28 64.241.88.17    4294967294    100      0 6347 i
*>i209.83.163.128/26 209.83.160.22   4294967294    100      0 6347 i
*>i209.83.182.208/28 209.83.160.22   4294967294    100      0 6347 i
*>i209.102.112.4/30 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.112.8/30 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.112.208/28 209.44.160.105  4294967294    100      0 6347 15131
?
*>i209.102.112.232/30 209.44.160.105  4294967294    100      0 6347 15131
?
*>i209.102.112.240/30 209.44.160.105  4294967294    100      0 6347 15131
?
*>i209.102.112.244/30 209.44.160.105  4294967294    100      0 6347 15131
?
*>i209.102.112.248/30 209.44.160.105  4294967294    100      0 6347 15131
?
*>i209.102.118.8/30 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.118.12/30 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.118.16/29 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.118.24/29 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.118.32/28 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.102.118.48/28 209.44.160.105  4294967294    100      0 6347 15131 ?
*>i209.126.143.192/27 209.44.160.105  4294967294    100      0 6347 i
*>i209.144.220.128/25 209.44.160.105  4294967294    100      0 6347 i
*>i209.223.197.128/27 209.83.160.22   4294967294    100      0 6347 i
*>i209.242.13.152/29 64.241.88.17    4294967294    100      0 6347 10692
13950 i
*>i209.242.13.204/30 64.241.88.17    4294967294    100      0 6347 10692
13950 i
*>i209.242.13.212/30 64.241.88.17    4294967294    100      0 6347 10692
13950 i
*>i209.242.13.216/30 64.241.88.17    4294967294    100      0 6347 10692
13950 ?
*>i209.242.13.220/30 64.241.88.17    4294967294    100      0 6347 10692
13950 ?

On Sun, 17 Dec 2000, Daniel L. Golding wrote:

Travis,

By doing a summary-only aggregate, you can lose routing information that
your downstreams want seen by the global internet. A good example of this
is prepending. If I only advertise a /14, then supress a /24 that is
subordinate to that block, I may fail to advertise a prepend upon that
/24 block. Paying customer don't like stuff like that.

BTW, ARIN is pretty clear that it's allocation policies are NOT intended
for use as filtering criteria. Most folks seem to get along fine,
filtering at the /24 level. It's not like most core routers at large ISPs
are 7500s with 64mb anymore. 

- Daniel Golding







Current thread: