Cell Relay Archive

[Marrone Pablo <marronep@telefonica.com.ar>]

In article <854584344.8563@dejanews.com>, manfredi@arl.bna.boeing.com
writes:
[snip]

>> I think they are orthogonal to any discussion about connection-orientation
>> (seems like I'm discussing ... never mind). ABR guarantees a minimum cell
> >rate, while UBR does not. UBR is more similar to regular LAN service.




And then murray@pa.dec.com stated


>Time for an alternate opinion.

>UBR is Unspecified Bit Rate.  The system doesn't promise anything.
>This is frequently called Useless Bit Rate.

>ABR does not specify a minimum cell rate.  That's a different dimension
>alltogether.  ABR is Available Bit Rate.  It requires some sort of
>feedback so that the client can figure out how much bandwidth is
>available and adjust its transmission behavior to avoid lost cells.

>Implementing ABR is not simple.  There are two competing camps - rate
>based and credit based.


According to ITU T Rec I.371 (may 1996) ABR do specifies MCR (minimun
cell
rate), wich is the resource reserved for the connection asking for this
transfer capability. I believe the same happens in Traffic 4.0 by ATM
Forum.
Feedback is accomplished by means of OAM RM cells. ABR specifications
are now
stable, and standard implementations are expected during this year.


>> > 4)Which traffic (ABR or UBR) is Best-effort traffic?
>>
>> ATM is best effort. So is Ethernet. If the destination node suddenly goes
>>down, neither ATM nor Ethernet will provide a clue to the source. Not until
>>timers time out or higher protocol layers get involved.


>If you accept my previous comments then that should be UBR rather than ABR.


To my understanding, both traffics are grouped under the "best effort"
name
in many books, equipment manuals and management systems.


>Calling Ethernet "best effort" in this context is misleading.  A congested
>ATM switch will silently drop (UBR) cells.  When an Ethernet packet
>encounters a collision, the transmitting host knows immediately and
>can schedule a retransmission (after backoff).  The result from a
>users viewpoint is that Ethernet packets generally get through but
>the timing is hard to predict as compared to ATM packets that just
>don't get delivered.
>
>ABR acts like Ethernet - the packet will (normally) get delivered, but
>you don't know how long it will take.



ABR will also drop cells "silently"  in case congestion occurs, but CLR
value
is assured. CD and CDV are not controlled in ABR traffic.




---------------------------------------------
Pablo S. Marrone
marronep@telefonica.com.ar
-----------------------------------------------