Propagation Delay and Its Relationship to Maximum Cable Length

http://www.wildpackets.com/resources/compendium/ethernet/propagation_delay

by andrelaszlo • 13 years ago

78 48 13 years ago

andrewvc• 13 years ago

Cool article. Though it only talks about ancient Ethernet variants. Issues like packet collision don't happen on most networks because we use switches instead of vampire taps and hubs.

Can someone more knowledgeable explain the parts that are still correct?

X-Istence• 13 years ago

All of them. Just because Station A and Station B are switched means that we could now say that the Switch is Station C. So lets say Station A starts transmitting at the same time as Station C (our switch is sending a packet from Station B to Station A) we can still get a collision.

We no longer get a collision between the nodes on the switch because of the fact that the switch will buffer packets as necessary before sending them on to their intended target, which reduces the amount of collisions that can/do occur...

mikeash• 13 years ago

I thought this stuff was full-duplex now, with each direction using different wires. So there's only ever one device that can transmit on a given set of wires, thus no chance for collisions. Is that not the case?

thedufer• 13 years ago

Very misleading article title. The entire thing is about collisions, which only limit the length between ethernet cards, which is 5 times the length in the title. I suspect the 500m is related to the bandwidth and signal degradation rather than anything in the ethernet protocol.

tlb• 13 years ago

None of this applies to modern gigabit ethernet installations, which are full-duplex and switched so that collisions don't occur. Maximum cable length is limited by signal degradation, so the limit is soft. With good quality cables you can go a little beyond the rated maximum.

ajross2• 13 years ago

Exactly. The switches buffer packets. The article is talking about a CSMA limitation of the old unbuffered hub architecture (and the "real electrical bus" coax networks before that) which have been irrelevant for over a decade now.

And my experience is that you can actually get well beyond the official lengths for most signal types even with comparatively cheap cables as long as you're willing to test and throw out the odd cable or connector. Standards are written with conservatism in mind.

jcr2• 13 years ago

> Exactly. The switches buffer packets.

In most cases, you're right. The vast majority of switches use a sore-and-forward design (buffering), but when you get into latency-critical applications, another type of switch design is used, typically called a cross-bar switch.

http://www.researchgate.net/publication/220055485_A_low-late...

http://www.google.com/patents/US5339396

I think opinions vary on whether or not crossbar designs "buffer" but it mostly depends on how you define buffering. None the less, the low-latency crossbar designs try to eliminate the latency caused by buffering in the older but more common store-and-forward designs.

sliverstorm• 13 years ago

With good quality cables you can go a little beyond the rated maximum

This sort of thing is designed with a very, very comfortable margin. I'd expect if you wanted, you could get quite a lot beyond the rated maximum.

vineel• 13 years ago

You seem to have a lot of knowledge about this. If I wanted to acquire similar knowledge, where would I go? I know embarrassingly little about networks.

smithian2• 13 years ago

Not the OP, but I would pick up a copy of Ethernet: the Definitive Guide. Then move on to TCP/IP Illustrated, 1&2.

jacquesm5• 13 years ago

Cat6 and cat5 are recommended to go no further than 90 meters:

http://en.wikipedia.org/wiki/Category_6_cable

http://en.wikipedia.org/wiki/Category_5_cable

Links are full duplex nowadays, the bus architecture is long gone and hubs are rare (usually just old), switches are dirt cheap so no reason for hubs.

The limiting factor is simply the ability of the electronics to recover the signal reliably. You can go way beyond the specs if you are willing to accept ever increasing packet losses due to degradation of the signal, one way to improve throughput over such links is by shortening your MTU.

Another simple solution to make longer runs is to use a fibre-optic repeater (for instance this ancient 10 mbit device: http://www.blackbox.com/resource/files/productdetails/10542.... but there are tons of others, more modern gear: http://www.signamax.com/mediaconverters/13-10100basettx-to-1... as an additional advantage, fibre optics don't suffer from electrical interference and act as an optical barrier for electrical signals which is very handy in industrial environments).

Or to forego cabling entirely and to use wireless with directional antennae.

zalew3• 13 years ago

> Or to forego cabling entirely and to use wireless with directional antennae.

but then comes the weather factor.

jacquesm5• 13 years ago

Over links up to a few kilometers this is usually not a problem if you can get a line-of-sight setup.

drone• 13 years ago

We use a 6Mb point-to-point link via dish to a tower 2 miles away, and have never had a problem with weather - and we get some pretty serious storms here on the gulf coast.

Sami_Lehtinen• 13 years ago

I used any random cables to build network at one location. It worked very well, except when you turned lights on, there switch disco. Digital communication works, until it doesn't. Without testing it's hard to say if it works or not. Cable, environment, connections and network adapters, switches all affect the overall result. Working doesn't mean it would be required to follow specs.

jallmann• 13 years ago

> switches are dirt cheap so no reason for hubs

Hubs are nice to packet-capture from a small device without a managed switch in between, so it's a bit of a bummer that they're hard to find these days.

jacquesm5• 13 years ago

Yes, that's a good trick. And splicing the RX pair to two different cards only works in rare cases, usually you have to lock the ports to a certain rate/duplex mode.

ghshephard2• 13 years ago

Most cheap switches these days will let you do port mirroring - that is, taking one or more ports, and "mirroring" the data into another "Mirror Port" (which you can then run wireshark against)

jpdoctor• 13 years ago

> Since we know that Ethernet operates at 10Mbps or 10,000,000 bits per second

I'm trying to figure out if this essay was written in the late 80s or the early 90s.

powertower• 13 years ago

> Electrical signals in a copper wire travel at approximately 2/3 the speed of light.

An interesting fact - in those copper wires (or any type of wires) the electron velocity itself is less than 1 centimeter / second (and often in the mm/s range). Whether it's DC or AC, it does not matter much.

Think of the wire as a tube with a bunch of ping-pong balls inside - filling it up tightly from start to end. If I stick my finger into one end of the tube, a ball will come out at the other end almost instantaneously (regardless of how long the tube is), but the balls themselves only moved a small distance - and only at the speed/velocity I pushed my finger at. But the "signal" "traveled" at a much higher velocity.

So in DC - electrons only travel a few centimeters / second (not anywhere near light speed), and in AC - you're pushing/pulling the same electron back and forth for eternity at about the same speed as it travels in DC.

Now it's of course a bit more complicated (the electrons rattle around at much greater speeds, and the mentioned velocity is a net average), and there are some edge cases, but for the most part the above is accurate.

I'm just not 100% sure of the speed of electrons in a superconducting loop (Cooper-pair electrons) - if they are truly relativistic or not.

And if this does not blow your mind, you should not be reading about electricity. PHDs in the electrical field don't even know this stuff.

http://en.wikipedia.org/wiki/Drift_velocity

TL;DR; If you raced a snail and an electron in a wire together, the snail would win - https://www.youtube.com/watch?v=jbi7gJTPSXk

greenmountin• 13 years ago

One of the coolest things about superconductors is "kinetic inductance"[1]. There is no energy in a normal current. As you say, the electrons don't travel all that fast most of the time. They feel a field, accelerate until they bump into something and start all over again. That's resistance; the energy stored by an inductor or capacitor (a current or voltage) is in the magnetic or electric field nearby.

This isn't true for Cooper-pairs in superconductors. They go fast enough that there is true energy stored in their velocity; there's nothing to stop them, they're a superfluid. This is used for certain types of detectors (MKIDs), and means that there is a bit of extra L in superconducting transmission lines, above the geometric inductance.

[1] http://en.wikipedia.org/wiki/Kinetic_inductance

weichi2• 13 years ago

It's not really the case normal metal transmission lines don't have some "extra" L, is it? The magnetic field still penetrates into any real metal, so you have some extra inductance.

But I agree that it's very important to understand that a superconductor is not the same thing as a "perfect conductor", and the kinetic inductance is a great example of this. The ideas behind MKIDs are very cool.

scoot2• 13 years ago

So if my tube of ping-pong balls is infinitely long, am I sending my signal infinitely fast?

Myrth2• 13 years ago

You're forgetting about propagation of force within the ping-pong ball. It is not immediate.

pm902• 13 years ago

No signal can travel faster than light. If you put pressure sensors along the infinitely long tube, you'll see that the signal reaches the points at different times.

We're just so used to seeing it happen instantaneously that it's slightly difficult to imagine otherwise

gmaslov• 13 years ago

The signal would only travel at the speed of sound in ping-pong balls. What that is precisely is very difficult to estimate. Try "wave propagation in granular materials" for a Googleable phrase. Even the case of a perfectly aligned 1-D chain of ideal ping-pong balls is not straightforward to analyze; the confinement force probably has a significant effect. (disclaimer: not a physicist)

paxswill• 13 years ago

Nope. Imageine instead of ping-pong balls you had a blue squash ball (TIL: the color is important! [0]). They're squishy, and so when you poke the ball at your end, in reality you're causing it to visibly deform. It then pushes back out to a sphere, moving away from you. In doing so, it pushed into the next ball. This process of deform->reform->...deform->reform takes time, and that time is the propagation delay.

[0] http://en.wikipedia.org/wiki/Squash_(sport)#Playing_equipmen...

Myrth2• 13 years ago

What blows my mind is that PhDs don't know this stuff.

I have only high school level ed, but I had to train many CS grads for real-life positions, since they were absolutely not fit for job. High schoolers were grasping the ideas much better.

pm902• 13 years ago

You know, I thought a lot about this. In India, there is a weird phenomena that I noticed: many of the best teachers want to teach, not college students, but high school students preparing for the rigorous engineering entrance examinations. Of course, it helps that they get paid a lot more than university professors, but the ones that taught me, for instance, had a genuine desire to impart knowledge where it would matter the most. They loved it when we would solve a difficult physics/math problem. By contrast, they complained that most college students were too distracted with other things. I found this to be true when I went to college, when, of a class of about ~500 students, only about 50 were actually honestly learning.

My hypothesis is that many college grads don't make the most of their easy access to learning, but have an incredible sense of self-entitlement. Those in high school are still at a very impressionable age; curious about the way the world works and hungry to find out more.

gcr• 13 years ago

A PhD gives you a specialty, not a well-rounded education. That's why it exists.

CS PhD students are not electrical engineering PhD students.

jleader• 13 years ago

When was this article written? The last time I worked in an office with 10Base2 cabling (aka "thin ethernet", using RG-58 coax about 5mm thick with BNC connectors) was 1995. The last time I saw 10Base5 (aka "thick ethernet", using an extra-stiff variant of RG-8 coax about 10mm thick, in continuous runs with "vampire taps") was sometime in the late 1980s.

And if the maximum cable length were simply proportional to signal propagation speed / bit rate, then why does 10Base2 have a shorter maximum length than 10Base5? They're both carrying 1e6 bits per second. Do signals really travel 3 times faster through fat coax than through thin?

smoyer• 13 years ago

Propagation delay is dependent upon a variety of factors and while the numbers the article quotes may be correct for thick-net and thin-net, the cable we use for CATV trunk-line is significantly faster at about 0.85c (http://www.china-satellite.com/sdp/1385167/4/pd-5853918/8382...).

Interestingly, the propagation delay through single-mode fiber tends to be significantly lower than through copper (well ... again it depends on the cable), but the 0.66c speed is a pretty good number for fiber.

ghshephard2• 13 years ago

"You may know that the minimum frame size in an Ethernet network is 64 bytes or 512 bits, including the 32 bit CRC. You may also know that the maximum length of an Ethernet cable segment is 500 meters for 10BASE5 thick cabling and 185 meters for 10BASE2 thin cabling. It is, however, a much less well known fact that these two specifications are directly related."

Odd, given that the VERY first thing anybody is taught about CSMA/CD ethernet is how the maximum length of cable and minimum packet size are precisely related. Indeed, it's almost the only thing people are taught about the relationship of those two.

weichi2• 13 years ago

Interesting article. A nit that can be picked is that it is not generally true that electrical signals in copper travel slower than c. The reason the signal travels slower than c in an ethernet cable is that the copper is surrounded by dielectric material, and it is the presence of the dielectric that slows down the signal. If the 10BASE5 coax cable was filled with air instead of foam, the signal would travel at c.

Propagation Delay and Its Relationship to Maximum Cable Length

Related Stories