From: Jim McCoy <mccoy@ccwf.cc.utexas.edu>
Date: Sun, 7 Nov 93 15:18:13 PST
To: pierre@shell.portal.com (Pierre Uszynski)
Subject: Re: some newbie DC-net questions
In-Reply-To: <9311070412.AA14773@jobe.shell.portal.com>
Message-ID: <199311072317.AA05272@flubber.cc.utexas.edu>
MIME-Version: 1.0
Content-Type: text


pierre@shell.portal.com (Pierre Uszynski) writes:
> > From: Jim McCoy <mccoy@ccwf.cc.utexas.edu>
> > [regarding collisions on a dc-net and detection of them]
> 
> Actually, it seems to me undetected collisions are not as likely as this
> makes it appear: Every person trying to transmit is monitoring
> at the same time to make sure the message they transmit does appear in
> the "sum of differences".

Yes and no.  In theory undetected collisions are unlikely.  In practice you
need to design against them in certain areas because unless you are talking
to someone on a true broadcast medium (ethernet, etc but not any internet
protocols above the datalink layer...) you are going to have to fake the
broadcast and depending on how you do this you could waste a significant
chunk of time and bandwidth passing a "token" around until the collision is
detected.  There will always be a delay in the time between broadcast and
reception so certain operations (like defining your channel in a
multiplexed broadcast network or getting any token or other identifier
necesary to let others know that someone is talking and they should not try
to transmit, etc) need to include steps to make sure that collisions are
detected early before significant effort and bandwidth is involved, where a
collision could make an entire round of messages need to be repeated...

> When you detected that your message collided, typically, you stop
> transmitting. You then decide on a random time delay, wait for that
> duration and try again if the way is clear (all zeroes carrier).

And make sure that your backdown includes a bit of random wait added on or
else two colliding speakers will constantly run into each other.  A good
example of how to deisgn and layout such a system can be found in CSMA/CD
networks like enet (IEEE 802.3 i think...) or the satellite broadcasting
networks, etc.  The methods necessary to make such a system work most
efficiently have already been designed by others for networks that share
many common characteristics with a dc network, makes sense to use them, no?
:)


jim