[ntp:hackers] ntp and signalled IO

Poul-Henning Kamp phk at phk.freebsd.dk
Tue Apr 18 17:07:25 UTC 2006


In message <2078.1145379681 at sa.vix.com>, Paul Vixie writes:

>if SO_TIMESTAMP means the kernel can put a timestamp on an i/o event, and
>it includes serial i/o, then i'm all for it.

SO_TIMESTAMP is only for (UDP) sockets.

For logical signals like 1 Hz from GPS and atoms we have the PPS-API
(RFC2783).

Timing serial data has no API at this point in time and I doubt there
will be much call for it.

>but eventlib won't make any timings worse, it'll just make the code better.

Well, it might result in a different order of the event scheduling
compared to a hand-rolled loop and thus impact userland timestamping
of packet reception.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk at FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.


More information about the hackers mailing list