Ethernet Questions

Factual Questions
1

I have some questions regarding Ethernet.

  1. How are Ethernet and TCP/IP linked? I know that TCP/IP is a
    protocol,
    but what is Ethernet if it is not a topology or a protocol?
    Can Ethernet use a protocol other than TCP/IP?

  2. Can an Ethernet network have a topology other than a star topology?

  3. Ethernet switches use CSMA/CD to ensure data can be transmitted
    successfully.
    If each computer has a separate connection to the switch why does the
    computer have to listen
    before it can transmit and why would it need to detect if there is only
    one computer using each cable?

  4. Does Ethernet guarantee data transfer at 100 or 1000 megabits per
    second? Would each computer still get
    100 megabits per second download speed if one computer was transferring
    a large file or if all computers connected to the switch were
    downloading alot of data at once would it slow down? Is it its speed which means that Ethernet is the most popular? Is it suitable for all types of files including large ones?
    Are there any other advantages or disadvantages of it?

Thanks for answering these questions.

2

  1. Ethernet refers both to the networking hardware used to make Ethernet networks, and to the protocol that runs on that hardware. This protocol is of a lower level than TCP/IP, so TCP/IP is said to run “on top of” Ethernet. Other protocols can run on top of Ethernet besides TCP/IP.

  2. Yes. In fact, a pure star topology is unusual in Ethernet. Each Ethernet circuit can support a large number of machines simultaneously. A small Ethernet network may only have one network segment, connected by a hub.

  3. CSMA/CD is used when several computers share the same circuit (as with a hub). Routers, as you correctly deduce, do not need to do collission detection between seperate circuits. The collission detection is done by the network interface hardware. So every computer on an Ethernet network will do collission detection to ensure that its messages to not colide with others on the same network segment.

  4. There are many different Ethernet standards of varying bandwidth rates. The most common is 100 megabit Ethernet, and 1000 megabit (aka gigabit) Ethernet is becoming more and more common. Gigabit uses the same category 5 cabling as 100 megabit Ethernet, but is guaranteed over shorter distances, and of course requires network cards that can transmit and receive that fast. 100Mb or 1Gb are theoretical saturated bandwidth limits and are rarely reached in a simple communication between two points.

Ethernet is certainly popular for its speed, though there are plenty of other network technologies at there that are as fast, and faster. Ethernet is popular also because it’s an open standard and very easy to set up and maintain. The type of data being transmitted doesn’t matter much; that’s more a question for the kind of protocols you’re running on top of the Ethernet.

3

As someone who’s written (most of) a device driver for an ethernet adapter, I ought to be well palced to answer this:

They’re related the same way as highways are related to vehicular trafic. Ethernet is a way for one computer to send a short message to another, but offers no way of finding out where the other computer is, how to get there, or making sure that the messaage arrives. It’s a complex matter, and I can’t explain it here.

Yes! In fact, (10/100-base-T) Ethernet is logically a ring network, although it’s physically a star. Old thin-ethernet (using coax) actually had a physical star topology.

Because logically the topology is that of a ring network. (I must admit that I’ve never seen the initialism CSMA before, but I guess it has something to do with collision detection.) Simple hubs broadcast all data. Independent of destination address, all incoming packets get transmitted to all ports. (There are some weird half-hubs / semi-switches who keep track of who’s where and send garbage to the other ports, but they’re (thankfully) rare today.)

No! Ethernet has a theoretical throughput of 10/100/1000Mbits per second - Only if there are no colliions, no contention, no congestion.
I could go on, but I haven’t got the time right now. If you want to know anything specific, just ask.

4

#1: Ethernet’s the data-link layer protocol that regulates the transmission of packets over the LAN. TCP/IP is more of a transport and network layer protocol that deals with the making of connections for message sending. (UDP is connectionless).

The upshot of this is that you can run TCP/IP on whatever kind of LAN or WAN you like- Ethernet, token-ring, FDDI, etc… it deals with higher level messages.

#2 and 3: Actually, Ethernet’s LOGICALLY a bus topology. Hence the carrier-sense multiple access, collision detection(CSMA-CD) part. Originally, it developed from the ALOHA packet-radio network, in which all stations shared a certain frequency band and they had to determine if anyone else was transmitting, otherwise they’d collide.

Once into the physical cabling world, this translated into all the stations sharing one wire. Original 10base-2 ethernet had terminators, etc… because they all shared one wire. The PHYSICAL star topology is more of a convenience thing- the hub just takes the place of the one shared wire, but ethernet works the same anyway.

4: If there was one sending station and one receiving station, then yeah, you’d get the rated speed. In more real-world situations, you get roughly the bandwidth divided by the number of stations as your effective bandwidth for a station.

This is why they have switched ethernet, which gives line-speed to each port on the switch (I don’t think it does CSMA-CD anymore exactly) and other topologies such as rings (token-ring and FDDI), which are better under heavy load than Ethernet.

The simplest answer as to why Ethernet is so common is that it’s by far the cheapest. Go look around on the web for 100BaseT cards and hubs vs. token ring cards and hubs. You’ll find that there’s no contest. Gigabit stuff is pretty expensive, but that’s mostly datacenter backbone type stuff, due to teh extremely short drop lengths you can have.

5

Here’s some pages that show the “official” OSI network layers model. Ethernet/token ring/modems are at the very bottom. TCP/IP is up a little higher (and straddles layers, which is why other layer models will have fewer than OSI).

It is really important to note bump’s point that Ethernet is a bus techology. Not a star, not a ring. A bus. You can add in extra stuff to make it physically a star, but it still thinks it’s a bus.

You can avoid CS/CD only if: there are two devices on the bus and they are duplexed (which can be done with twisted pair but not coax). That is a very limited situation and the hardware for such a specialization would end up costing more even though it did less. Economy of scale.

Ethernet was vieing with broadband networking systems at the time. In those systems, the idea was to send many bytes, modulated, over several different frequencies. But the multiplexing/demultiplexing hardware for doing this was much more expensive at the time. (It’s gotten a lot cheaper since, but still not as cheap as Ethernet.) It did take Ethernet a while to catch on. Many people thought that such a simple, collision prone, system wouldn’t work in real life. I remember reading the usual “Ethernet is doomed” articles when it was commercially announced. Laugh out loud, I had been using it for 5 years already and loved it.

It’s cheap, it’s simple, it’s expandable, it works.

Here’s a short article on Ethernet history.

6

All kinds of fun nitty-gritty Ethernet stuff can be found here:

http://www.erg.abdn.ac.uk/users/gorry/course/lan-pages/enet.html

7

I agree with most of your points (I wuvs me some Ethernet), but I’m not sure I agree with the one above. Isn’t the situation you describe basically the situation as it is between two full-duplexed switch ports or a full-duplex capable NIC connected to a switch port ? Such hardware is seriously cheap these days. (Care must be taken to avoid full-duplex/half-duplex mismatch. That’ll work well enough for a quick low-load ping test and fail miserably under load conditions.)

Well, that’s only true for copper. With the proper GBICs, you can shoot Gigabit through 70 km (43.5 miles) of Single-Mode Fiber. (Of course, digging 43.5 miles of ditch for the cable is a royal pain. But it would work…)

8

Clarification on Ethernet performance: an Ethernet switch will have a backplane fast enough to handle full-duplex data transmission at maximum speed to every port on the switch simultaneously. For example, a 16-port 100mbit Fast Ethernet switch will have a 3200mbps, or 400MB/sec, backplane to handle 200mbit to and from each of the 16 ports. An Ethernet hub, however, copies all data transfers to all attached systems. As multiple systems transfer data, performance quickly drops to unacceptable levels due to packet collisions and the small amount of total bandwidth available.

So, in short, if you’re using an Ethernet switch, you’re going to get the biggest share of the available bandwidth that both you and the host are capable of delivering. If three other people are leeching porn from Bob, you’re gonna get at most 25mbps. If you’re the only one leeching from Bob, but you’re leeching porn from three other people, then once again you’re only going to get 25mbps. However, if you’re leeching porn from Bob, and Bob is leeching porn from you, you’re both going to get full 100mbit speeds, thanks to the miracle of full-duplex data transmission. These examples ignore whether your respective systems can handle the I/O load of simultaneous transfers. On 100mbit Fast Ethernet this isn’t really an issue, with GigE it’s the biggest one.

A note on what friedo said about GigE: Gigabit Ethernet will not run on normal CAT5 cable, it requires CAT5E(nhanced) or CAT6 cable for proper operation.