Ethernet auto-negotiate time

Sep 12, 2004 3 Replies

Hi, I have been trying to calculate how long a successful Gigabit Ethernet auto-negotiation sequence takes and the time I have calculated seems to be quite high. From referring to the 802.3 standard:



Time [average] between consecutive FLPs: 16ms Gigabit Ethernet uses 4 LCWs to auto-negotiate: one Base Page, one message Next Page and two 1000BaseT unformatted Next pages.



Time to send four FLP: 4 X 16ms = 64ms


802.3 standards specify that each LCW should be received 3 times before acknowledgement. Therefore, 4 FLPs [3 X LCW + 1 acknowledgement LCW] are required for exchanging 1 LCW.

Therefore, minimum time to auto-negotiate 1Gbps link speed = 64 ms X 4 = 256ms



Considering that Ethernet is a LAN protocol, this seems to be too large a vaue. Could somebody please tell me where I have made a mistake? Thanks!



Chamara Gunaratne snipped-for-privacy@csee.usf.edu University of South Florida



No mistake, your numbers are right. Moreover, this is the *minimum* time for a successful negotiation; in the event that an FLP burst is corrupted, it may require more than 3 repetitions before acknowledgment (but this is considered a rare event).

Given that Auto-Negotiation occurs just once at link initialization, it is not considered a big deal to take a few hundred milliseconds. In practice, this time overlaps with other events, such as machine boot sequences, power-up self tests, etc., all of which take even longer.

By the way, the minimum negotiation time is less for lower data rates, since the Base Page alone conveys all the necessary information to negotiate 10/100 Mb/s operation.

-- Rich Seifert Networks and Communications Consulting 21885 Bear Creek Way (408) 395-5700 Los Gatos, CA 95033 (408) 395-1966 FAX

Send replies to: usenet at richseifert dot com

Thank you. I was not sure about the calculated time and am glad to know that it is OK.

Gigabit Ethernet is now common in desktop computers and we find that NIC power consumption is about 4% ~ 5% of the overall idle power consumption of the computer. And with copper 10GigE NICs expected within a couple of years that percentage is expected to increase significantly. At the University of South Florida my advisor, Dr. Ken Christensen and myself are studying the feasibility of dynamically changing the link speed and hence reducing the power consumption of the NIC. And the time taken to change link speed is an important metric.

Thanks again for your kind response,

Chamara Gunaratne

I would think you could minimize the amount of circuitry that had to be clocked at full speed during idle time.

For battery powered devices, though, I agree you want it as low as reasonably possible.

-- glen

Join the Discussion

Have something to add? Share your thoughts — no account required.

Didn't find your answer?

Ask the community — no account required