# Shortest ethernet frame time = 5.76uS?

• posted

Could someone check my calculations? I'm looking for the shortest time an Ethernet frame will occupy the wire. I make it about 5uS on 100Mbit as follows.

Minimum frame size in bytes is 46 payload + 14 header + 4 FCS = 64. Preamble (still needed on full duplex) of 8 octet times Total of 72 octets * 8 bits = 576 bits. Each bit time is 1/100,000,000 seconds. (IIRC the signalling rate is

125,000,000 bits with a 4B/5B encoding so we do get 100,000,000 data bits in a second.)

Finally 576 * 1/100,000,000 = 5.76uS

On 10Mbit I make the figure ten times as much 57.6uS On Gbit I think they kept the basic timing budget from 100Mbit so make it's smallest-frame time about the same as for 100Mbit.

Corrections welcome. Does the above look about right?

-- James

• posted

At one point on the cable or over the legal length of cable?

• posted

Umm, well if the speed of signal propagation over copper cable is about 0.7 times the speed of light (not sure of the exact figure but that's about right, isn't it?) we have the speed as

0.7 * 300,000,000 metres per second = 210,000,000

If that's right then in 1uS the signal would propagate 210 metres so I think the length of the cable becomes almost insignificant - up to a standard length of 100m anyway. Maybe it would add 0.5uS to the total, over that length.

Incidentally I think the old co-ax thinnet relied on reflected wave signalling so signals had to go to a terminator and bounce back for each bit - so the whole cable became energised for that bit long before the next bit was put on the wire.

Just some more comments that people may be able to shoot down... Still I don't mind being the clay pigeon. It would be good to get some of these ideas confirmed ... or denied.

-- James

• posted

Oh, yes. Also without looking it up I think you are right. Something about the slot time being needed to allow collisions to be recognised so of course not being needed on full dulpex. Thanks for pointing this out.

• posted

"insignificant" depends on what's behind the question. Can you tell us what you are trying to do, or what's concerning you so the question can be answered more appropritely.

• posted

James Harris wrote: (snip)

Without looking it up, I believe that carrier extension (to keep the frame time up) on gigabit is only done on half duplex links.

-- glen

• posted

I mean the signal propagation delay (absolute maximum 0.5uS) appears to be "almost insignificant" compared with the minimum frame time of

5.76uS. There's no application relevant. I just wanted to check my figures. Of course the resulting thread may be useful for someone to refer back to.

-- James

• posted

Don't forget the minimum interframe gap of 96 bit times, making the total 6.72 us. Granted, there is no *frame transmission* during this time, but it still must be enforced, so the time cannot be used for anything else.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

• posted

[snip]

The cable length is irrelevant here. The frame takes 6.72 us (see my earlier post for the correction to the 5.76 us figure above) regardless of where you "look" at it. Even if the cable was 10,000 km long, it is still a 6.72 us frame at any point on the cable; no device can "see" the cable at any point other that the sole one at which it is connected.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

• posted

note 100 Mbps still supports half duplex and "repeaters" - the repeater can effectively eat some bits at the beginning on the packet.

The practical effect is that the number of bits recieved can be less than the number you are supposed to send - and the flip side to that is you could lose a few in the sedning device and the packet still notionally conforms to the spec at the reciever.

note if you add tags then the minimum grows by 4 bytes per tag for

802.1q tags unless tags are optional in your setup (or stacked MPLS labels if you use those).

finally you are meant to impose an "inter frame gap" between packets (96 bit times) - so if you measure from start of 1 min sized frame to another one dropped in directly behind it there are 12 more bytes of overhead to worry about.

So - it sort of depends on what and where you measure....

Nope - no practical repeaters on GigE, so no need for frame stretching.

• posted

FWIW for half duplex at 100 Mbps you are allowed 400m of fibre - which is the limit for 2 minimum size packets to be injected at the 2 ends and to "overlap" at both ends - ie you can detect a collision consistently.

with copper cables there is some delay thru a 100 Mbps repeater, so you are down to 100m of UTP on each port - which is why repeaters are much less flexible for 100 Mbps Ethernet (but they got replaced by switches fairly quickly, so it didnt turn out to be a big problem).

• posted

OK. Thanks.

• posted

You are confusing propagation delay with transmission time. The transmission time is constant (6.72 us) regardless of the length of the cable on which it is impressed. Granted, the frame won't get to the other end of the cable until the propagation delay has elapsed, but that propagation delay is not "added" to the transmission time, since the sending station can "re-use" the (now quiescent) cable at its location even while the signal is still propagating to the far end.

Not at all. In fact, terminators were placed at the ends to *prevent* excessive reflected signals.

The cable acts as a delay-line memory. It literally "stores" the frame as it propagates down the transmission line. A long cable can hold quite a bit of data. For example, a 5000 meter 1000BASE-X single-mode fiber has a propagation delay of approximately 25 us; that's over 37 minimum-length Ethernet frames (or 2 *maximum* length frames!) in the pipeline, sent but not yet received at the far end.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

• posted

James Harris wrote: (snip, I wrote)

As far as I know, no repeaters were ever sold. I don't know if NICs have the ability to run half duplex even if not connected to a repeater.

-- glen

• posted

Rich Seifert wrote: (snip)

I remember reading some time ago about an undersea fiber optic cable with a propagation time of about 0.1s. The bit rate is probably pretty high, too.

-- glen

• posted

Stephen wrote: (snip)

The 100m limit for copper is mostly attenuation. It is there even for full duplex. For 100baseTX with two class II repeaters, the total cable length limit is 205m, which, if you allow 100m out to each port leaves 5m in between.

For fiber, the velocity is a little different, and the attenuation limits are a lot different.

-- glen

• posted

The 0.1s delay is the right order of magnitude, representing about

20,000 km of fiber. At 1 Gb/s, the cable can store 12.5 MBytes of data!

Note that a sending station using TCP would have to have buffers at least TWICE that large, and typically much larger, because it takes over

0.2s (the round-trip time) before receiving an ACK for transmitted data.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 228-0803 FAX

Send replies to: usenet at richseifert dot com

• posted

Just for completeness.

Regarding commercially produced 1000BaseTX repeaters.

The Cisco GigaStack GBIC may have been a commercially available GBE Repeater. Allbeit with only two ports each.