Actually, for a graphical study of this, check out this:
http://nic.crc.ca/bgp/bgp-all.html
It's part of what Geoff Huston has developed recently to illustrate such
issues (albeit in some way).
On Thu, 22 Nov 2001, Gert Doering wrote:
> Hi,
>
> On Thu, Nov 22, 2001 at 01:46:35PM -0800, Michael K. Smith wrote:
> > >> *>i24.154.32.0/24 213.239.22.241 200 0 8918 701 7046 ?
> > >> *>i24.154.33.0/24 213.239.22.241 200 0 8918 701 7046 ?
> > >> *>i24.154.34.0/24 213.239.22.241 200 0 8918 701 7046 ?
> > >> *>i24.154.35.0/24 213.239.22.241 200 0 8918 701 7046 ?
> [..]
> > > Even if 7046 has multiple links to 701, it would be possible to send
> > > out a /20 *and* all the /24s, and tag the /24s as "do not export
> > > elsewhere".
> >
> > What if 7046 has a connection to 1239 as well?
>
> They don't, at least not in any way that is visible on the looking glasses
> or in our BGP table.
>
> We're not talking "what-if's" here. We're talking "people are garbaging
> the whole world" here.
>
> > If 701 announces the /20 as
> > such, while 1239 announces the /24, traffic is going to follow the longest
> > path every time.
>
> A clear case for "aggregation in 1239 as well".
>
> > > this means: 73,000 prefixes are there that are more specific than what the
> > > registries allocate, read: "that are likely to be aggregateable to a LOT
> > > less".
> >
> > That's because lots of ISP's have multihomed customers using a
> > sub-allocation of their address space.
>
> You cut the interesting paragraph: I didn't say that it's possible to
> remove all of those 73000 prefixes. I estimated about 30000, and this
> leaves lots of room for "multihomed sub-allocations".
>
> Currently there are about 10,000 AS numbers in use. Can you give me a good
> reason why this means 105,000 prefixes? Those "multihomed customers"
> usually have *one* address block, meaning *one* prefix per AS.
>
> > I am not arguing that there are misconfigured routers out there where people
> > couldn't figure out the "mask" statement, but there are also lots of small
> > providers trying to provide quality of service to their small markets by
> > having more than one upstream provider.
>
> This is not a way to provide quality of service to anybody.
>
> Announcing poorly aggregated bullshit is a good way of providing very bad
> quality of service. People have enough of this and WILL start filtering.
>
> We already DO filter "more specific than /24". Saves about 2000 prefixes.
>
> gert
> --
> USENET is *not* the non-clickable part of WWW!
> //www.muc.de/~gert/
> Gert Doering - Munich, Germany gert@greenie.muc.de
> fax: +49-89-35655025 gert.doering@physik.tu-muenchen.de
>
wfms
This archive was generated by hypermail 2b29 : Sun Aug 04 2002 - 04:13:24 EDT