Merit Network
Can't find what you're looking for? Search the Mail Archives.
  About Merit   Services   Network   Resources & Support   Network Research   News   Events   Home

Discussion Communities: Merit Network Email List Archives

North American Network Operators Group

Date Prev | Date Next | Date Index | Thread Index | Author Index | Historical

56k modems/frame/ptp revisited (LONG)

  • From: Tim Wolfe
  • Date: Tue Sep 01 11:45:07 1998

13 people so far have asked me to post a summary to the list on this
subject, so here it is. I have included all responses I received, 
with the name of the respondent at the top of their answer.  Both 
questions are included at the top for continuity purposes.

Question #1:

Anyone have a nice layman's terms explanation of the throughput/latency
differences between a 56k modem, a 56k leased line, and a 56k frame relay
circuit?  This is assuming a "perfect" lab enviroment, ie. the modem
connects and sustains at maximum capability, the frame cloud is not
overloaded etc..  Please reply off list.

Question #2:

I've gotten lot's of responses on my original question, but I'm finding
perhaps I didn't ask the correct one.  What I really need to know is what is
the real world effect of latency on throughput?  I know that frame relay has
a higher latency than a point to point line, and that a point to point is
generally regarded as faster.  Would someone mind explaining the specifics
of that relationship a little bit?  Again, off list please, unless someone
thinks it's an appropriate topic for this list..

Here are the responses I received for both questions in no particular order.

David Newman wrote:

"56K" modems really top out at 53.3 kbit/s because of an FCC regulation (I
believe Canada has a similar rule). They theoretically can hit 56 kbit/s
outside North America, though tests have showed 51-52 kbit/s to be the
upper end for products shipping today.

Absent congestion both a frame PVC and a leased line really do run at 56

David R. Dick wrote:

"56K" modems take advantage of the digital nature of much of the
present phone system.  Unfortunately, they can only do this in
one direction, so you get 56K down from the Internet to the user
(actually only 53K, or so, because of electrical power limitations
in the signalling method) and 33K from the user to the Internet.

56K leased lines are point-to-point connections.  The data travels
more or less directly (in electrical terms, anyway) between the
user and the Internet -- 56K in each direction.

56K frame relay is a switched service and two point-to-point
connections are involved: one from the user to the switch
and one from the switch to the Internet.  You get 56K in each
direction, but there is some delay (latency) because of the
diversion through the switch.  In some places there may even
be more than one switch involved, with extra latency, as you
might expect.

Paul G. Donner wrote:

Basically, you'd need to look at the framing in each case as well
as any Layer 2 acknowledgement mechanism that might be implemented.
You can get this from a number of textbooks and put together an
explanation yourself if you are so inclined.

This would only consider up to the L2 overhead on the links (if
applicable -> e.g. Frame Relay) which is probably what you are
asking for.  These link types are effectively in different
categories and it would be sort of like comparing apples to

Chris Cappuccio wrote:

56K is the low end of the leased line spectrum, with 56K leased line and
56K frame relay performing almost identically if you are using similar
framing protocols.  You will the top 56K performance from these... Lowest
latency (20-40ms)

56K modem is at most 53K because of modem-enforced FCC limits, and because
of the analog-digital conversion, you will get high latency (100-200ms)
and lower throughput then a leased line.

David R. Dick wrote:

The simplest explanation involves the fact that most networking
is TCP (not UDP) so packets travel in both directions, with
acknowledgements, etc.  When an ACK is delayed because of
latency, the movement of the real data packets is also delayed.
So it will just plain take longer to deliver the same data

Karl Auerbach wrote:

As for so-called 56K modems -- they aren't.

They are only 56K if the FCC allows, which they don't, so top speed is about

And then it's only 56K in the direction from an ISP who has a digital
ISDN based modem towards your "analog" 56K modem.  The return path is
33Kbits (or worse.)

If you take a couple of off the shelf "56K" modems and hook 'em up
over a normal POTS line you'll end up with about 33Kbits (or worse)
full duplex.

I learned all this stuff a few weeks ago when I hooked a couple of
routers together and found that my investment in 56K modems was wasted
-- the best one can get is 33Kbits full duplex unless I invested in an
ISDN link at one end and got the 56K ISP-end gear.

As far as I know only 3COM makes a single BRI ISP-end 56K modem.  It
runs about $250.  Usually they come in larger packs to hook to an ISDN PRI.

Chris Cappuccio wrote:

Latency and throughput are NOT related in this case!

The speed of light itself puts a limit on how fast we can transfer data
globally with fiber optics.  Bigger pipes will simply ensure that you can
get more data through at the same time.  Once you have maxed out speed
with the most efficient protocol, you are limited by the speed of light ;0

Anyways, the latency introduced with a modem connection is much larger
then that of a dedicated line because of the two analog<->digital
conversions.  All-digital connections like ISDN, FR, and PTP connections
are very low latency.

Paul G. Donner wrote:

FR is effectively frame/packet switched.  PT-PT circuits do not have
frame/packet structures (the transport may use T1 type framing but
this is different).  The major latency affect in FR is due to buffering
of the packet/frames and and switching them.  PT-PT don't go through
this process.  So effectively both types FR and PT-PT have propagation
delay (e.g. speed of light limitations) but FR exhibits also processing
delays which are neglible or non-existant with PT-PT circuits.

Robert Gourley wrote:

Given that there in no congestion in the frame relay network and
that both the 56Kbps DDS line and the 56Kbpx frame relay PVC follow
exactly the same route and have the same number of route miles,
then the only difference in the latency of the two circuits is
the effect of two serialization delays (one at the ingress and
one at the egress) in the frame relay example.  Serialization delay
is = (packet size in bits)/(56,000 bits/sec).

Blaine Christian wrote:

Since you want to assume ideal conditions you will remove a significant
difference between the 56k Modem and leased line.  You will find that
under ideal conditions both of these should perform equally well.  Since
a leased line/modem relies on less telco equipment it should peform
better than the frame circuit.  A 56k frame circuit will increase your
latency by virtue of being processed by several frame switches.  If the
frame switches have to tag frames they may introduce more latency into
your circuit (even if they do not tag your frames).

The difference between all of these should be negligible (depending on
switches and the amount of equipment traversed) and throughput should be
equal depending on the equipment you are using (of course).

	Tim Wolfe, Network Administrator
	ClipperNet Internet Access Services	541.431.3360 voice 541.431.1176 fax
   ...Across the Desktop and around the World...	 

Discussion Communities

About Merit | Services | Network | Resources & Support | Network Research
News | Events | Contact | Site Map | Merit Network Home

Merit Network, Inc.