[time-nuts] Timing on Ethernet

[Pablo Alvarez Sanchez]

Dear timing colleges, 

At CERN we are considering the possibility of using Ethernet as a real
time field bus. 
We may use IEEE 1588 to distribute precise UTC and Ethernet Powerlink or
a similar home made product to guarantee real time. 

In order to have full control over the network we are thinking of
designing our own hardware using FPGAs (including the physical
interface). It looks like we will end by implementing 100Base-TX or
10Base-T rather than gigabit Ethernet, just because it looks much more
robust and simpler to design. 

I am just a beginner in the Ethernet world. I have the following
questions concerning the timing: 

1) Two ways calibration performed in IEEE 1588 needs a symmetric path
between nodes. In a presentation made by 'Timing Solutions', they say
there is a 10% of asymmetry on the path for 100Base-TX. I guess this
comes from the fact of using one pair for the go and a different pair
for the return path.
Do you have some figures for the two ways asymmetry on a single pair? 

http://ieee1588.nist.gov/2006%20IEEE1588%20Agenda/Stein_Sub-nanosecond%2
01588%20final.pdf


2) I was thinking of locking a PLL to the Ethernet carrier in order to
minimize the jitter between modules. Is there a continuous clock or
encoded signal between frames? If there is nothing specified between
frames do you think one can send some kind of clock or may that provoke
some perturbations on a standard  PHY?

Thanks for you attention

Pablo 

-------------------------
Pablo ALVAREZ SANCHEZ
CERN - AB Department
CH-1211 Geneva 23

GSM   : +41 (0)76 48 72191
Phone : +41 (0)22 76 78431 
Fax   : +41 (0)22 76 69318
Building : 864-1-A30