> > I'm doing all my testing with ping -s to set the size of the packets.
When
> > I set packet sizes significantly higher than the MTU size I start
getting
> > lost packets. I was hoping that defragmentin the packets on node B will
> > cause less collisions because all of the fragments will be received
before
> > transmission begins.
>
> Even if you could 'solve' this problem, I think ping is a rather
artificial
> way to test the setup if you're normally going to be using TCP for real
> communications (which I assume will be the case).

Yeah, that's probably true. The network performance is acceptable as I guess
TCP backs off the packet rate until fairly reliable communications occours.
However, occasionaly I am getting long periods of no connectivity (sometime
up to a couple of minutes or so). I'm not sure weather this is related to
the collisions though.

> TCP will overlap the packets, so A will send packet 1, then packet 2, then
> packet 3, etc... whilst waiting for the acknowledgement of packet 1.   It
> won't send, wait for ack, send, wait for ack.....   Therefore I think
you'll
> still run into the same problems as you're seeing now, which I also think
are
> due to collisions on the wireless network (because only one out of A, B
and C
> can transmit at any one time).
>
> How far apart are A, B and C - specifically A and C ?   Can they pick up
each
> other's radio signals, or is it possible that A is talking to B, C can't
tell
> because it can't hear A, so C starts talking to B as well, and B loses
> everything ?

The nodes are actually quite far apart (I'm using high gain antennas and am
actually getting around 50% signal strength). A is 1.2km from B and C is
another 200m from B. B is at the top of a hill. A and C cannot reach each
other directly (at least not with enough signal strength for a packet to
pass through without error). A and C are also on different subnetworks. B
routes packets between the two networks.

Thanks for your comments,

Russell


Reply via email to