Talk:5-4-3 rule

Copyright violation?

This article seems a lot like

• http://www.answers.com/topic/ethernet-way-versus-ieee-way

so which came first? Maybe Answers borrowed from this article? This needs research. --DThomsen8 (talk) 13:59, 12 January 2010 (UTC)

Answers.com is a wikipedia mirror. The bottom of the page identifies the material as coming from Wikipedia. -- Whpq (talk) 14:25, 12 January 2010 (UTC)

Rationale

The following uncited rationale for the 5-4-3 rul has been supplied (twice) by user:96.51.205.25

A repeater normally listens for the 0101 preamble and then locks onto the bit stream. Once locked on it would then repeat each bit out the other port(s). However some number of bits would be lost at the start while the repeater was locking onto the bit stream. As the frame propagated through each repeater the preamble would get shorter and shorter. Too many bits lost meant that NIC cards may not have enough preamble bits to lock on and the entire frame would be missed.

As each manufacturer of repeaters (hubs) did slightly different implementations each repeater operated differently. Each different repeater would lose more or less bits while locking on. Some would lose 1 or 2 while others could lose as many as 5 or 6 bits. You could create a network with more repeaters if you knew how each repeater operated. However this information was not always easy to obtain and difficult for users to calculate. Hence the 5-4-3 rules was created. It was easy to understand and worked in all cases. In a lab at DEC they knew how many bits their repeaters would lose and knowing this were able to create an 11 segment, 10 repeater, 3 active segment (11-10-3) network that was completely viable.

Clause 9.5.5.1 of IEEE 802.3-2008 describes repeater requirements and doesn't appear to allow shortening of the preamble. Here's another ref that supports a requirement to transmit a full preamble: [1]. A couple other refs are less clear and may support the claims above: [2], [3].

I believe the primary purpose of the rule is to insure collision detection works properly. ~KvnG 15:35, 2 June 2013 (UTC)

The preamble shortening by repeaters does happen but usually it's not an issue and therefore it gets ignored, the collision detection being a far more important point.

The preamble actually is longer than necessarily required – different hardware needs different number of cycles to sync – and any repeater will lose the cycles it required to sync and can't make them up since the frame is following right behind. After some research I've found a source for this in the 802.3 standard and will add it to the page in a minute. Zac67 (talk) 16:32, 2 June 2013 (UTC)

Repeaters also decrease IFG (Inter Frame Gap), which is also a limiting case. As well as I know, repeaters regenerate the preamble, buffering the data for a some bits. This can decrease IFG. Gah4 (talk) 20:24, 27 July 2017 (UTC)

Wrong

This page may or may not be right in the case of coaxial and FOIRL links, but isn't at all close for 10baseT.

It would probably be better to just delete the page. Gah4 (talk) 02:20, 18 May 2015 (UTC)

The 5-4-3 rule is a rule of thumb, nothing more, nothing less. If you've got constructive, sourced changes to include go right ahead and improve the page. --Zac67 (talk) 17:25, 21 May 2015 (UTC)

One question is why this page exists. I suspect that if anyone now installs coaxial ethernet, they install only one segment. The part of the rule that says "populated segment" means "coaxial segment", and link segment meant (at the time) FOIRL. Collision detection is much faster on optical and UTP segments, and as far as the rule goes, UTP segments count as link segments. But even more, and the description doesn't make this obvious, coaxial segments count as populated even with only two (the repeater) taps. It is nice to have for historical reference purposes, but that has to be in the context of the technology when the rule was written. Gah4 (talk) 20:32, 27 July 2017 (UTC)

Nobody installs new coax Ethernet any more, it's long obsolete. A link segment can be anything (10M), check Clauses 13.3 and 1.4.255. --Zac67 (talk) 06:29, 28 July 2017 (UTC)

I suspect that once in a while, someone connects two devices with a coax segment, likely 10base2, to keep some old device running. I doubt anyone connects two or three coaxial segments through repeaters. From 13.2, mixing segments are 10base5, 10base2, and 10baseFP, while link segments are FOIRL, 10baseT, 10baseFB and 10baseFL. Link segments have separate transmit and receive paths (fiber or UTP) and detect collisions by a signal on the receive lines at the same time as one is transmitting. 10baseFP is the fiber equivalent of coaxial ethernet, which allows more than two hosts on a single fiber segment, and has collision detect similar to coaxial. (I don't know that it ever caught on.) Gah4 (talk) 07:27, 28 July 2017 (UTC)

Duh! I hadn't caught the "full-duplex medium connection" in 1.4.255 before – apparently you're right! (Obviously, it's not a full-duplex link as in switched Ethernet, but a full-duplex medium as in -Fx and -T – no coax, except for 10BROAD36). --Zac67 (talk) 11:18, 28 July 2017 (UTC)

This used to be popular discussion in comp.dcom.lans.ethernet, including posts from the person who designed coaxial ethernet. The one I didn't know until yesterday is 10baseFP, which is a shared medium optical fiber. In the case of shared medium, it takes longer to detect a collision, and that adds to the total round trip time. Also, using model 2 rules, you find that with an all 10baseT network you can have five repeaters, and even 150m segments (with Cat 5 cable) and still be legal. Five is good, as it allows for a central repeater, connected to others in the same wiring closet, and then additional repeaters at each end station (office). Well, by now the center should be a switch, but for at least a short time that was important to know. Gah4 (talk) 18:44, 28 July 2017 (UTC)

link segment

Note that 10baseT is properly listed as link segment, but also is not normally used just to connect repeaters together, but most often to connect end stations to repeaters. The rule was formulated when repeaters were very expensive, so needed to be minimized. As technology changed, the rule did not keep up with the changes. Gah4 (talk) 19:54, 10 March 2019 (UTC)

It's about the full-duplex capability - on an FDX medium collision detection is much faster than on a HDX-only, shared wire. 10BASE-T is fine as a link segment and it was used that way, at least by me. This is all historical now, I don't think anyone is using a repeated 10M segment any more... --Zac67 (talk) 21:49, 10 March 2019 (UTC)

Well, the statement applies to 10baseT repeater or not. Specifically, it implies that they are only used for repeater to repeater links, when we know that they are used for more. I suspect that there are still a few small 10baseT repeaters around, used where speed isn't important, or where no-one bothered to replace them. Also, some devices have a built-in repeater for pass-through use. (Some might have a switch, though.) Gah4 (talk) 00:43, 11 March 2019 (UTC)

10BASE-T can be used as a link segment, whether there are more nodes attached to either repeater hub or not. I've never seen integrated repeaters, but it would apply there as well. Again, it's about the full-duplex capability of the PHY. --Zac67 (talk) 06:58, 11 March 2019 (UTC)

I'm surprised to see any mention of full-duplex here at all. The rule was created for coaxial shared medium - inherently half-duplex stuff. A full-duplex link doesn't resolve collisions promptly, it eliminates the potential for collisions. ~Kvng (talk) 12:39, 14 March 2019 (UTC)

The idea isn't that it is full duplex with data, but that the actual link has separate wires that go each direction. Collision detection, for FOIRL and (half duplex) 10baseT, is done by sensing bits on the receive port while transmitting. Contrast to coaxial Ethernet where the transmitter must measure the average voltage on the cable, through an appropriate filter, and find it higher than an appropriate threshold. The faster collision detect changes the round-trip collision detection delay. It changes it enough that a pure (half duplex) 10baseT network can have six segments with five repeaters, and even up to 150m for each segment. Five repeaters is significant, as it allows for a central repeater, connected in a wiring closet to many repeaters, and an additional repeater at each end station (office). At that time, bridges (now called switches) were very expensive, and so minimized. Gah4 (talk) 17:18, 14 March 2019 (UTC)

Gah4's 100% correct. The rule was created for coax mixing segments interconnected with FOIRL or 10BASE-FL link segments which are full-duplex media, enabling a much faster collision detection. The same applies to 10BASE-T. The communication itself can never be full duplex as we're talking about repeaters here. --Zac67 (talk) 20:31, 14 March 2019 (UTC)

OK, it may have enough wires to do so but but hub-connected 10BASE-T does not provide full-duplex communications. We need to find a different way to describe this. I will take a crack. ~Kvng (talk) 15:46, 17 March 2019 (UTC)

The number of wires is irrelevant. Repeater (hubs) can't ever provide full-duplex communication, it's about a full-duplex medium that enables faster collision detection. --Zac67 (talk) 17:24, 17 March 2019 (UTC)

I have made some proposed changes. In addition to avoiding associating full duplex with repeater-hub-based 10BASE-T, I have removed the uncited claims that collision detection is faster on these systems. Collision detection time is limited in all systems by propagation time through the media. The speed advantage of 10BASE-T is not due to separate transmit and receive medium but due to the role of the hub plays in detecting collisions and the star topology which imposes shorter limits on the maximum distance between the farthest stations. These same advantages can theoretically be achieved with a single (half-duplex) medium. ~Kvng (talk) 14:23, 20 March 2019 (UTC)

It seems that my favorite books, written by the person who did much of this, is not in the references. The timings needed to figure it out are in the model 2 rules, which are in the references, but not a link to the actual rules. As the appropriate 802.3 documents are freely available, they should be linked. Propagation time is a large part of collision timing, but repeater delays are also important, and a big part of repeater delay is detection of an actual collision. Gah4 (talk) 18:24, 20 March 2019 (UTC)

Kvng, this is all in IEEE 802.3 Clause 13 which is an article source – we don't need a cite after each sentence, do we? You're wrong about the "speed advantage of 10BASE-T": collisions are simply detected faster on full-duplex media. With 10BASE-FL, -T, FOIRL, ... (with duplex media), a collision is sensed by the first bit of the carrier on the receive channel. With 10BASE5/2, there's no separate receive channel (simplex media) and a collision is sensed by "illegal" voltage levels due to constructive interference – obviously this takes a few bits to happen and eats into the time budget for collision detection, hence shortening the collision domain. --Zac67 (talk) 19:00, 20 March 2019 (UTC)

It is more than a few bits, as it is done by filtering the voltage on the cable, and then sensing it as more (absolute value) than a specific value. It is complicated by the need, in doing receive mode detection (for repeaters) to differentiate one nearby sender from two at the far end of the cable. The tolerance on collision detection is pretty tight (precision resistors needed). Model 2 has the official timings used for conforming implementations, with bit budget for each type of interface. We should have a reference to it. Gah4 (talk) 20:51, 20 March 2019 (UTC)

Point taken on the extra bit times taken to detect a collision based on carrier magnitude. It looks like 9 bit times are allocated to this so 10BASE-T diameter can be 1.8% bigger based on this factor. I stand by my edits though. We do have "prompt collision detection" with both systems. ~Kvng (talk) 16:36, 23 March 2019 (UTC)

I don't know if there is a standard meaning for prompt. It is important in nuclear fission, as prompt fission is what makes bombs work. Coaxial ethernet collision detect is enough slower compared to 10baseT, that you can have six segments with five repeaters for 10baseT, instead of the five segments that you might otherwise suspect. Gah4 (talk) 07:20, 26 March 2019 (UTC)