flight TCP hack? (was: Re: sources for GL demo programs)

[Vernon Schryver]

In article <24904 at uflorida.cis.ufl.EDU>, kml at mosquito.cis.ufl.edu (Kevin Lahey) writes:
> I've always been a bit disappointed that the dogfight program could only
> be used on my local net, since it uses broadcast UDP packets. ...


With luck, the SGI IGMP expert will answer, urging everyone not to panic,
since dog now knows about UDP multicast, has an officially registered multicast
port, and the IGMP multicast router is standard in 3.3.  SGI once again has
a single, coporate circus because of it.  The circus had been fragmented
since dog switched from XNS multicast to UDP broadcast.

If you were at Interop, you might have noticed people in the NOC and the
Interop booth dogfighting with each other and with people in the SGI booth,
despite the fact that widely separated networks were involved.  Our expert
had created an IGMP tunneling router on one of the IRIS's.

If someone happened to set up such tunneling routers on the Internet, or if
IGMP gets popular in the Internet, people with direct connections to the
Internet will be able see how many dog fighters are required to saturate
the NFS backbone.  I did my part by twisting arms for IGMP at router
vendors, who said "wait until OSPF."

Is dogfighting between academic institutions a "fair use?"  Will dog work
with the packet delays in the Internet?  Sounds like "research" to me.



Vernon Schryver,   vjs at sgi.com


P.S.  Since the SGI Internet gateway is too paranoid to IP forward, we will
not be among you lucky bad guys filling up the Internet.

P.P.S.  Remember, IRIS's don't kill networks, people kill networks.