[ntp:hackers] Further to the timestamping issue

David L. Mills mills at udel.edu
Sat Jun 14 21:31:50 UTC 2008


Guys,

Not to bring up a dead horse, but I don't think everybody was completely 
happy about the timestamp capture issue; at least, I have few comments 
about the white paper Stuart Venters and I prepared. However, and mainly 
to clear the air with a practical, worked-out scheme, I have updated the 
NTP On-Wire Protocols briefing at 
http://www.eecis.udel.edu/~mills/database/brief/onwire/onwire.ppt and 
white paper http://www.eecis.udel.edu/~mills/onwire.html to fix a few 
types, cleanup the presentation and add an interleaved broadcast mode. 
For yucks I included a short description of the NASA/JPL Proximity-I 
timestamping scheme used on Mars orbiters and landers. Cool.

For reference, the symmetric/broadcast interleaved protocol is backwards 
campatible with the present protocol, resists lost, duplicate and bogus 
packets and automatically adapts to basic and interleaved protocol 
modes. It has been thoroughly tested in simulation for both normal and 
error conditions.

If I can figure a way to capture TSC/PCC contents in the Ethernet 
driver, this puppy will bark. I've peeked at the FreeBSD Ethernet device 
driver for the Digital Tulip (if_de.c) and found it vaguely readable. 
I'm not a device driver guru by any means, , although I feel like an 
elephant in a china shop. The idea is to append the TSC to the end of 
the receive and transmit buffer when the frame is complete. Crude, but 
this is an expereiment, not a specification proposal. I have a couple of 
machines with Broadcom NICs that the if_de.c driver can handle.

Comments and/or guffaws humbly accepted.

Dave


More information about the hackers mailing list