[ntp:questions] NTP Hardware Support

David Garijo dgmangas at live.com
Mon Jun 22 20:54:53 UTC 2015

The project isn't really for fun, it's more like a "change of course" of another project.A fully functional NTP service has been implemented on a FPGA. It's way faster, but it isn't standard at all.
Right now, I'm trying to use Linux to standarize the whole thing, so that's the reason 'cause I'm asking all this.I'm also trying to understand the similarities NTP and PTP, so maybe I can generalise this hardware support and design it for both protocols.
> From: stenn at ntp.org
> To: dgmangas at live.com
> CC: questions at lists.ntp.org
> Subject: Re: [ntp:questions] NTP Hardware Support
> Date: Mon, 22 Jun 2015 18:13:20 +0000
> David Garijo writes:
> > Hello,
> > 
> > I've been thinking of the possibility of adding hardware support for
> > NTP so that it can timestamp the packets sent and received more
> > acurately.
> > 
> > An example of this would be to overwrite the timestamp of the packets
> > (by hardware) right before sending/receiving them with the actual time
> > of departure/arrival if a "hardware support" option is enabled (like
> > some kind of timestamp unit).
> > 
> > It sounds as if I would need to modify the project heavily, so I
> > suppose the best and most realistic approach would be to add a new
> > option to the (Linux) kernel instead of rewriting the ntp programs
> > offered.
> > 
> > The code for the NTP daemon (ntpd) is quite complex, so I'm not quite
> > sure how nor when the packets are sent/received/timestamped. On the
> > other hand, as far as I've noticed, sntp uses libevent and
> > recvfrom/sendto to handl e the packets, so I tend to think that maybe
> > I should start by taking a look at sys/socket.h instead of worrying
> > about anything ntp related. Any suggestions? Do you know of an ideal
> > file/function that I should edit?
> Changing the timestamps as the packet leaves is not a good general
> solution, as there can be checksums involved.
> NTP already has some hardware timestamping support in it.
> I think the better approach is to track the actual hardware timestamps
> and use "interleave" mode.
> Is this a project you are doing "for fun"?
> H

More information about the questions mailing list