CSMACD Question

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

First off, I'm not a network person, I'm just taking a course, but my
background is wireless communications, so I'm more famliar with Aloha
than Ethernet.

At any rate, I'm having a debate with someone over the behaviour of
the contention algorithm and I'm wondering if anyone has a way to
break the tie (with some balck and white proof).

Question:

Take a 4 users network (A,B,C and D). A and B send and collide, they
then sit back and wait for their timers to go off to try again. C and
D then send and collide, they too set their timers and wait to try
again.

Point of Contention:

Do both A and B detect the collision of C and D, and if so, are they
then forced to reset their timers?

My view point is No, and here are my reasoning. Technically I think
they could detect the collision of C and D, but if this is the case,
then why do transmitters need the loopback of the Tx signal to the Rx
to compare what's Rx's if anyone (i.e. A or B) can detect the
collision?

Any literature I've found so far only deals with the transmitter and
receiver, none specifically detail the requirements of the passive
terminal. Another reason why I don't think this is the case as I
would expect it explicitly written that ALL terminals must reset their
timers.

Thanks,
Darren

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

D.K. wrote:

(snip)

> Question:

> Take a 4 users network (A,B,C and D). A and B send and collide, they
> then sit back and wait for their timers to go off to try again. C and
> D then send and collide, they too set their timers and wait to try
> again.

> Point of Contention:

> Do both A and B detect the collision of C and D, and if so, are they
> then forced to reset their timers?

> My view point is No, and here are my reasoning. Technically I think
> they could detect the collision of C and D, but if this is the case,
> then why do transmitters need the loopback of the Tx signal to the Rx
> to compare what's Rx's if anyone (i.e. A or B) can detect the
> collision?

Repeaters need to detect receive mode collisions (even
when they aren't transmitting), but ordinary end stations
don't.

-- glen

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <da3707b7.0411150737.3de58f8f@posting.google.com>,
qn_42@hotmail.com (D.K.) wrote:

> Question:
>
> Take a 4 users network (A,B,C and D). A and B send and collide, they
> then sit back and wait for their timers to go off to try again. C and
> D then send and collide, they too set their timers and wait to try
> again.
>
> Point of Contention:
>
> Do both A and B detect the collision of C and D, and if so, are they
> then forced to reset their timers?
>

The IEEE-standard algorithm specifies that stations only adjust their
retransmission timers based on collisions to which they are a party.
Thus, while both A and B may be able to detect the fact that there was a
collision (not involving themselves), they will not use this information
to adjust their retransmission timers.

In the original (coaxial Ethernet) design, it could not always be
guaranteed that a listening station would always detect collisions
between 3rd-parties; thus, the algorithm was specifically designed not
to depend on that ability.

> My view point is No, and here are my reasoning. Technically I think
> they could detect the collision of C and D, but if this is the case,
> then why do transmitters need the loopback of the Tx signal to the Rx
> to compare what's Rx's if anyone (i.e. A or B) can detect the
> collision?
>

Transceivers do not detect collisions by comparing the transmitted data
to received data. The loopback of TX to RX came "naturally" in the
design of the original coaxial Ethernet. To maintain backwards
compatibility with existing controller silicon, we included this
capability in later 10BASE-T systems.

> Any literature I've found so far only deals with the transmitter and
> receiver, none specifically detail the requirements of the passive
> terminal. Another reason why I don't think this is the case as I
> would expect it explicitly written that ALL terminals must reset their
> timers.
>

We actually considered the behavioral impact of using so-called
"receive-based collision detection" in the original design, and
concluded that any performance improvements were minimal, and not worth
the added design complexity.


--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

In article <oDsmd.201$df.18979@tor-nn1.netcom.ca>,
darren_kuhn@Mitel.com wrote:

> In what way? Do repeaters have to detect the collision itself or just the
> jam signal?
>

"Jam signal" is a misnomer. There is absolutely no difference between
the signal transmitted for "jam" and the signal transmitted for ordinary
data. "Jam" simply means "continue sending for some additional time," to
ensure that all parties to the collision properly detect it.

> The way it's being presented to me says that any station that the first
> station to detect the collision sends out the jam signal. If this is the
> case, wouldn't then pretty much everyone detect it 'first' because of
> propagation delays?

Stations detect collision in a medium-specific manner (the method of
detection is different in coaxial vs. twisted-pair vs. fiber systems,
etc.), through circuitry in the transceiver. Every transmitting station,
upon detecting a collision, "jams", that is, continues to send data for
another 32 bit times, before ceasing transmission and rescheduling the
frame.

Of course, all of this is now somewhat moot, as the majority of devices
operate in full-duplex mode using switched connections. Full-duplex mode
eliminates the entire CSMA/CD algorithm: no collision detect, no jam, no
backoff, no retransmission. Any discussion of the details of CSMA/CD is
of purely historic interest.


--
Rich Seifert Networks and Communications Consulting
21885 Bear Creek Way
(408) 395-5700 Los Gatos, CA 95033
(408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

 

[Guest]

Archived from groups: comp.dcom.lans.ethernet (More info?)

Rich Seifert wrote:

(snip)

> We actually considered the behavioral impact of using so-called
> "receive-based collision detection" in the original design, and
> concluded that any performance improvements were minimal, and not worth
> the added design complexity.

There is a National Semiconductor application note describing
both transmit mode and receive mode collision detect for coaxial
cable ethernet.

It was my understanding that repeaters required receive mode
collision detect. In the application note they determine
the parameters for running ethernet on 75 ohm and 93 ohm coax,
and they do both modes separately.

Using external resistors on the 8392 they can properly
detect collisions on such cables.

-- glen

 

[CJ]

Archived from groups: comp.dcom.lans.ethernet (More info?)

They only detect a collision if their workstation sends and collides.



"D.K." <qn_42@hotmail.com> wrote in message
news:da3707b7.0411150737.3de58f8f@posting.google.com...
> First off, I'm not a network person, I'm just taking a course, but my
> background is wireless communications, so I'm more famliar with Aloha
> than Ethernet.
>
> At any rate, I'm having a debate with someone over the behaviour of
> the contention algorithm and I'm wondering if anyone has a way to
> break the tie (with some balck and white proof).
>
> Question:
>
> Take a 4 users network (A,B,C and D). A and B send and collide, they
> then sit back and wait for their timers to go off to try again. C and
> D then send and collide, they too set their timers and wait to try
> again.
>
> Point of Contention:
>
> Do both A and B detect the collision of C and D, and if so, are they
> then forced to reset their timers?
>
> My view point is No, and here are my reasoning. Technically I think
> they could detect the collision of C and D, but if this is the case,
> then why do transmitters need the loopback of the Tx signal to the Rx
> to compare what's Rx's if anyone (i.e. A or B) can detect the
> collision?
>
> Any literature I've found so far only deals with the transmitter and
> receiver, none specifically detail the requirements of the passive
> terminal. Another reason why I don't think this is the case as I
> would expect it explicitly written that ALL terminals must reset their
> timers.
>
> Thanks,
> Darren