Cell Relay Archive

["Robert K.M. Larsen" <RobertLarsen@msn.com>]

We haven't actually measured the real throughput, but it can easily be
worked out.  Unlike more traditional protocols like X.25 and TCP/IP, where
you have a framing protocol at level 2, you can get very close to (i.e.
within 1% of) 100% utilisation on a cell-relay trunk.  With X.25, say, you
might only get a maximum of 70% data utilisation on a trunk, the rest of
the bandwidth being taken up with link management overheads (i.e. framing).

So, with a 53-byte cell, there is a 5-byte header, giving 48 bytes for the
payload.  This gives just over 90% data throughput.  You should be able to
apply this to any ATM trunk, regardless of speed.  Another point to bear in
mind, is that from the supplier of the circuit's/megastream's point of
view, the trunk is running at (nearly) 100% utilisation constantly.  Even
if there is no data to transmit over the ATM trunk, empty cells will be
transmitted to "fill the gaps" in the user data.

Robert.

----------
>From: 	Henry Baker
Sent: 	06 December 1996 04:15
To: 	Robert Larsen
Subject: 	Re: Throughput of 25Mbps ATM?  of 45Mbps ATM?

(A copy of this message has also been posted to the following newsgroups:
comp.dcom.cell-relay)

In article <01bbe2e0$878654e0$834995c1@robertl>, "Robert K.M. Larsen"
<RobertLarsen@msn.com> wrote:

> We currently use a Stratacom IGX-based ATM core network (though we will
be
> adding BPX switches as well), and have experienced close to 100%
throughput
> on some of our ATM trunks (when a LAN device "swamped" the frame relay
> access connection).  I have measured the management overheads (i.e. high
> priority cells), and it has been negligible (less than 0.1% of the trunk
> bandwidth).

What is this actual measured throughput over a 25Mbps ATM link in bits/sec
or
bytes/sec -- e.g., for a file transfer or even a memory-to-memory byte
stream ??

Thanks in advance.




Robert K.M. Larsen <RobertLarsen@msn.com> wrote in article
<01bbe2e0$878654e0$834995c1@robertl>...
> We currently use a Stratacom IGX-based ATM core network (though we will
be
> adding BPX switches as well), and have experienced close to 100%
throughput
> on some of our ATM trunks (when a LAN device "swamped" the frame relay
> access connection).  I have measured the management overheads (i.e. high
> priority cells), and it has been negligible (less than 0.1% of the trunk
> bandwidth).
> 
> Any error "packets" (I presume you mean cells?) will be automatically
> dropped by the receiving end, and no retransmission will take place. 
> Dropped/errored information is left to the higher-level protocols (in our
> case, X.25, TCP/IP, IPX, etc.) to manage.
> 
> Hope this was useful.  Feel free to contact me if you need more info.
> 
> Rob.
> 
> Henry Baker <hbaker@netcom.com> wrote in article
> <hbaker-0212960013410001@10.0.2.1>...
> > Could anyone tell me what the real, measured _throughput_, net of all
> > overheads, that one can get through a 25Mbps ATM link?  What about a
> > 45Mbps link?  Can it be calculated as a straight percentage of the
> > nominal number, or is there some more complicated formula?
> > 
> > I'm interested in a single fixed-bandwidth stream on a link with a
> > relatively small error rate.
> > 
> > What if I wanted any error packets to be simply dropped on the floor
> > and ignored, rather than trying to be retransmitted?  Would this
> > make any difference on the throughput available?
> > 
> > Thanks in advance for any info on this problem.
> > 
>