A couple of points not brought up (or didn't see in my quick scan). Use of the telephone network for data is on a permissive basis, it is designed for analog voice traffic, no guarantee by the telco that you can obtain any specific rate out of a given phone line. Frame Relay services is a shared system. You can purchase a 56kbps access line and port, but the important parameter is the committed information rate you purchase from the carrier. Most of Sprint's frame relay business may have a 56kbps access line with ZERO committed information rate. Welcome to use it and can peak to 56kbps, but your actual performance may vary anywhere from 0-56kbps. Can purchase 56kbps CIR, but it costs extra, would need to then look at the economics of using a dedicated line instead. Unles your app is to mesh several locations together, then frame relay could be used to set up virtual connections to multiple locations usning one access line, an important economic point. Daniel T. Scott Chief Engineer NASA Integrated Services Network (http://www.nisn.nasa.gov) MSFC Information Systems Services Office ******NOTE NEW AREA CODE FOR MSFC ****** 256.544.0131 voice 256.544.2554 fax dan.scott@msfc.nasa.gov

---------- From: Tim Wolfe[SMTP:tim@clipper.net] Sent: Tuesday, September 01, 1998 10:17 AM To: nanog@merit.edu Cc: ClipperNet Technical Staff Subject: 56k modems/frame/ptp revisited (LONG)

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 kbit/s.

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 oranges.

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 payload.

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 53K.

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 tim@clipper.net 541.431.3360 voice 541.431.1176 fax ...Across the Desktop and around the World... ---------------------------------------------------