I ordered a an increase in my ADSL speed from 1 Mbps to 3 Mbps (up) for the
New Year; it was installed early in the morning of Jan 3. The effect on NTP
has been nearly miraculous. I posted the loopstats.20120110 file on
pastebin.com (user name CElliott), but in summary the time offset was W/I ±
0.5 milliseconds of zero the entire day, except for four brief excursions to
about -5, -3.5, -2.5 and -2.5 ms. Just as unusual, the frequency deviation
is an almost constant 29.5 PPM for the whole day.
The poll interval was set at a constant 6 = 64 seconds using
min and maxpoll. I tried letting it range free between 4 (=16 secs) and 12
(=4096 secs), but ntpd soon went to a polling interval of 12 and stayed
there, and the time offset increased greatly and stayed there also. I have
been working on an algorithm (Tian, Y.-C., & Gui, L. (2011). QoC Elastic
Scheduling for Real-Time Control Systems. Real-Time Systems, 47(6). doi:
10.1007/s11241-011-9133-x) that is supposed to produce the optimum minimal
control loop monitoring rate based on the observed error in each control
loop. I am using ntpd to test it. Once this algorithm decides that a loop
is in control (the error is small) it puts it on the slowest scan rate, but
will move it to a faster scan rate it errors re-appear. No real-life
process subject to random deviations will stay in control if it is not
monitored.
I bought the increased ADSL speed (an additional $6 per
month from Verizon) based on an article (Wallsten, S. (2011). What gets
measured gets done. Commun. ACM, 54(11), 26-28. doi:
10.1145/2018396.2018406) that showed that the US was 18th out of 30 major
industrialized nations in terms of broadband speed. The reason the US was
in the middle of the pack was not because the country did not have the
infrastructure, but because consumers refused to purchase the faster
offerings. Consumers almost always bought the slowest broadband speed
offered. Indeed, it is true that 1 Mbps is sufficient for email and an
occasional Wikepedia lookup. ISPs figured, Why buy more infrastructure if
consumers will not purchase what is already offered? So I decided to throw
myself into the breach based on purely nationalistic grounds. The average
page download time, based on a DIY benchmark that runs in a random interval
every half hour, has increased from about 43,000 Bps to about 110,000 Bps;
the hugely increased NTP performance was a welcome dividend.
Charles Elliott
_______________________________________________
questions mailing list
[email protected]
http://lists.ntp.org/listinfo/questions
