Cell Relay Archive

[fahad@cs.pitt.edu (Fahad A Hoymany)]

Craig Partridge (craigp@world.std.com) wrote:
: Hi Fahad:

: I understood you were talking about cut-through.  What I'm pointing out
: is that the IP specification doesn't allow cut-through to start until the full
: packet has been received, because you are not allowed to route the packet
: until after you do the length checks.

Craig, 
  That's true in traditional IP, but what about people like at Ipsilon and
elsewhere who talk about forwarding complete packets without any
intervention from the router!  At least cut-through forwarding (as
described earlier) allows the router to see the IP header.  So, yes, going
strictly by the IP specification, you may not gain much by going to ATM,
other than a higher throughput rate.

  I see a great advantage of cut-through when the source is slow (or the VC
rate requested by the source is slow).  Once the router knows where the
packet is heading, it no longer has to wait for the remining cells to
arrive -- when the cells do arrive they find a VC binding in place and can
be switched at the ATM layer.  That's mainly the idea of switch-borne
router (SBR) which we proposed in a paper at ICCCN
(http://www.cs.pitt.edu/~fahad/SBR.ps).

Cheers
Fahad