Re: WG ACTION: 2 weeks to discuss [LL62] Do not space probes randomly

    Date:        Fri, 7 May 2004 10:46:39 -0700
    From:        Stuart Cheshire <[email protected]>
    Message-ID:  <[email protected]>

  | Collisions do not cause loss on Ethernet. They did not on thick coax, 
  | thin coax, or on UTP.

Of course they don't - that was never the issue, once again, you're
raising and dismissing a strawman.

The problem is packet arrival and the ability of *every* host on the
LAN to handle a packet train of length N - every host because they're
all broadcast packets.   And the same train in the same order every time.

Typically there's at least one host on a LAN that has a fairly small
threshold for N, before packets are dropped - and usually the same
packets every time.    If that host just happens to be the one that
has the same address as the one being probed by the packet that is
lost (on reception, not on the wire) every time, then the protocol
fails.

As Christian said, the implementation already has some kind of RNG
(or PRNG), it has to to calculate the first random interval.  Thus
there's essentially no cost in using it every time (a few extra
calculations, a multiply & modulus usually).   There's a clear benefit,
you don't get the same packet train every time, different packets
get lost each retransmit.    That allows the protocol to work.

Get off this hobby horse, fine a better one.

kre