[ntp:hackers] Question about USB3... anyone played with it?

Greg Troxel gdt at ir.bbn.com
Wed Dec 11 21:24:50 UTC 2013


Harlan Stenn <stenn at ntp.org> writes:

> "Gary E. Miller" writes:
>> Greg Troxel <gdt at ir.bbn.com> wrote:
>> 
>> > I have a GR601-W, which seems to have a 1ms variation.
>> 
>> Yeah, can't call that horrible, that is better than you'll get
>> on GigE sitting next to a Stratum 1.
>
> Really?  I'm used to sub-msec (like under .25) offsets and jitter with
> GigE LANs and S1 servers.
>
> And to make sure I'm on the same page as y'all, what is "variation"?

@stenn:

Sorry, what I meant is that the latency from the actual 1 pps edge from
the ublox chip in the GR601-W until a NetBSD kernel takes a timestamp
due to DCD transition seems to be explainable by 200 us plus a uniform
random number from 0 to 1ms.  I don't have enough data to show that, and
I'm not claiming sample-sample uncorrelation.  Just that there's 1ms of
fuzz on pps samples obtained this way.

@list:
I agree that even with only 100 Mb/s Ethernet, the USB errors are
large.  But the USB errors are small compared to what one sees with the
massively asymmetric routes seen on today's Internet.  From BBN to MIT
packets take about 1.5 ms (I used to see 2.5 ms round trip times via
ntpq).  But the return packets go via chicago and take about 23 ms.
So that's just a cautionary tale not to dismiss USB PPS as useless, but
rather to be thoughtful about relative errors.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 194 bytes
Desc: not available
URL: <http://lists.ntp.org/pipermail/hackers/attachments/20131211/d0c71ea3/attachment.sig>


More information about the hackers mailing list