Livingston & BGP & multicast USENET (was blah blah blah)

Tue Jun 10 03:21:11 UTC 1997

bmanning at ISI.EDU writes:

> 	Livingston has serious problems with its BGP code and
> 	has since it was first available (going on six
>       months now)

I don't suppose you'd care to elucidate?  

I'd be curious about how many updates/second this box can
handle while forwarding across all interfaces at line
rate, and how quickly from the receipt of an update
switching to another path the forwarding engine begins
sending traffic along that path, and how many
christmas-tree packets flowing at line rates to a series
of changing destinations are lost or mis-forwarded.

Of course I tend to be interested in environments where
traffic is already enormously aggregated, where all or
nearly all the available bandwidth is being used, 
where dozens of BGP updates/second is commonplace, and
where occasionally there will be "flutter" that causes
large numbers of prefixes to attract data down oscillating
paths.

These are difficult conditions for any router, and
unfortunately nobody seems to have a public benchmark
which tests for them, other than the "it works/it doesn't"
in the middle of large pieces of the Internet.

I would be interested in actual "it works/it doesn't"
commentary (and possible explanation of "it doesn't", if
that's the case) involving this particular router from
some sizeable ISP or other.  Learning from vendor mistakes
is of use to many of us.

> > Will IP multicast help with the usenet stuff? 
> 
> 	Tangental question.

Kurt Lidl and company's excellent MUSE paper, presented at
USENIX in Winter 1994 details early attempts to distribute
news via the MBONE.

Unfortunately the MBONE is not IP multicast in any
serviceable or saleable sense, and that and the
work involved in making it difficult to forge multicast
USENET news tends to make the distribution of news slower
and less reliable than the parallel interactive unicast
scheme.

On the other hand, as there has been a great deal of work
on eliminating multiple MBONE tunnels across providers,
this both could be less true and less important for people
who are concerned about the number of copies of articles
moving across their networks.

On yet another hand, it might make sense for "proper"
multicast facilities to be deployed as supportable
products before jumping into doing MUSE or something like
it in a serious way.

On yet another hand, satellite dissemination of news, both
with pagesat-style distribution and sniffing at satellite
point-to-point transmissions using amusing tcpdump scripts or
programs directly using the berkeley packet filter, have
been happening for some time.

On the final hand, NNTP distribution is much less broken
than Web distribution, and the latter distribution problem
probably deserves to be attacked first.

Oh yah, out of curiosity, what native multicast support is
in the Livingston box, and what does it use for routing?
PIM?  DVMRP?  CBT?

	Sean.