Re: ntpq times out if NMEA refclock configured?


"Richard B. Gilbert" <rgilbert88@xxxxxxxxxxx> wrote in message
news:d72dnXEvvMabBfXZ4p2dnA@xxxxxxxxxxxxxx
R Jenkins wrote:

"Richard B. Gilbert" <rgilbert88@xxxxxxxxxxx> wrote in message
news:9sqdnZb36qXzoPrZRVn-sg@xxxxxxxxxxxxxx

R Jenkins wrote:

"Richard B. Gilbert" <rgilbert88@xxxxxxxxxxx> wrote in message
news:L7udnQ5X_9grBvvZRVn-vA@xxxxxxxxxxxxxx


R Jenkins wrote:



Hi,

I'm trying to add a GPS refclock to my server.
After total failure with a basic Trimble TSIP output GPS plus the
parse
clock, I'm now using a Garmin GPS25 and the NMEA refclock.


<big snip>

After rereading a little more carefully, I notice that your frequency
correct of -495.9 PPM is on the ragged edge of the 500 PPM limit. It
is
unusual for a clock to have a freqency error this large; most are below
50 PPM in absolute value.

Does your system have a kernel parameter called "HZ"? Is it set to a
value greater than 100? I believe I have seen references to values of
both 250 and 1000; neither value works well with NTPD. The system
seems
to lose clock interrupts when HZ is greater than 100. YMMV but if you
are not using 100, give it a try.


Hi,
thanks for the replies.

The -495.9 ppm seems to be a symptom of the refclock problem. Without
the
NMEA refclock it was -60 after a few minutes, long before it had settled
properly.
I think it does have a fast Hz setting (I've seen it somewhere but I
can't remember where or what it was set to..) However, it's a 3.2GHz
processor so I don't think it should struggle too much.


I have the PPS pulse set to 200mS.
The PC does not normally have a display, I use telnet (well, SSH) from
my
desk.
Running minicom at 4800 Baud with NTPD stopped shows the GPS serial data
is present:
$GPRMC,073153,A,5319.0516,N,00106.9355,W,000.0,000.0,140506,004.0,W*76
$GPRMC,073154,A,5319.0516,N,00106.9355,W,000.0,000.0,140506,004.0,W*71
$GPRMC,073155,A,5319.0516,N,00106.9355,W,000.0,000.0,140506,004.0,W*70
...
I'm not sure how to remotely monitor the DCD line.


Simply having the 'server 127.127.20.0 prefer' line in causes the ntpq
hang.

I've just got around to checking the log immediately after starting
ntpd:

May 14 08:19:51 gate2 ntpd[28723]: ntpd 4.2.0a@xxxxxxxx Sat May 13
10:39:48 BST 2006 (1)
May 14 08:19:51 gate2 ntpd[28723]: precision = 1.000 usec
May 14 08:19:51 gate2 ntpd[28723]: Listening on interface wildcard,
0.0.0.0#123
May 14 08:19:51 gate2 ntpd[28723]: Listening on interface wildcard,
::#123
May 14 08:19:51 gate2 ntpd[28723]: Listening on interface lo,
127.0.0.1#123
May 14 08:19:51 gate2 ntpd[28723]: Listening on interface eth0,
192.168.0.43#123
<Other interfaces trimmed>
May 14 08:19:51 gate2 ntpd[28723]: kernel time sync status 0040
May 14 08:19:51 gate2 ntpd[28723]: refclock_nmea: time_pps_kcbind
failed:
Invalid argument
May 14 08:19:52 gate2 ntpd[28723]: too many recvbufs allocated (40)

It looks like there is some problem with the kernel PPS interface, but I
have no idea what...
I used this patch:
PPSkit-light-alpha-3328m-2.6.15.1.diff
on a clean download of kernel 2.6.16.9 - there were a couple of rejects,
but they seemed to be pretty obvious & went in easily by hand..

I'm happy to try another (recent) 2.6 kernel if there is one with a
known
working patch?

Another test: Leaving the 'flag 3 1' out stops the refclock error line
in
the log.
The 'too many recvbufs allocated (40)' line seems to be triggered by the
NMEA refclock regardless of any other settings; it does not appear when
the NMEA clock is commented out in ntp.conf

Robert Jenkins.




If the HZ setting is causing the problem, it has little to do with
processor speed!! The problem seems to be that various device drivers
mask or disable interrupts for a period covering two or more clock
interrupts causing one or more to be lost with each occurrence.

The messages about "too many recvbufs allocated (40)" were associated
with
a bug in ntpd that I believe was fixed more than a year ago. You might
want to try the latest version of ntpd. You can download it from
http://ntp.isc.org/bin/view/Main/SoftwareDownloads


Hi,
I can understand the Hz setting messing up the accuracy, but I don't see
it
would stop things running altogether?
It was at 250Hz, I'm presently compiling a kernel with it at 100Hz to see
what effect this has.

I thought I had tried the latest dev release (as per my last post), but
it turns out I had two copies of ntpd in different locations.
Using the ./configure options from the Redhat source does not put ntpd
into the /usr/sbin directory as with their build, they must be patching
the paths somewhere as well.

Having properly cleared the old files & rebuilt again, I am getting
slightly better results, but it's still not locking to the GPS.

After around an hour:
# ntpq -c peers
remote refid st t when poll reach delay offset
jitter
==============================================================================
xGPS_NMEA(0) .GPS. 0 l 18 64 377 0.000 -482.37
160.720
+gate.jrw.intra 130.159.196.118 3 u 6 16 377 0.196 131.924
19.370
*mail.alsys.ro .GPS. 1 u 49 64 377 67.099 25.813
69.507
+cronos.cenam.mx .GPS. 1 u 24 64 377 217.026 -10.000
92.241

ntptime
ntp_gettime() returns code 0 (OK)
time c8121481.1f3f8000 Sun, May 14 2006 21:41:37.122, (.122063),
maximum error 603208 us, estimated error 39319 us
ntp_adjtime() returns code 0 (OK)
modes 0x0 (),
offset 30763.000 us, frequency -432.622 ppm, interval 4 s,
maximum error 603208 us, estimated error 39319 us,
status 0x1 (PLL),
time constant 2, precision 1.000 us, tolerance 496 ppm,
pps frequency -1.251 ppm, stability 0.000 ppm, jitter 0.000 us,
intervals 0, jitter exceeded 0, stability exceeded 0, errors 0.

Initially ntptime was giving code 5 (error) & showing frequency 0 ppm.
At this, the offsets stayed reasonably constant. I had deleted the drift
file before starting so it would not be upset by the previous problems.

After it changed to code 0 & status 0x1, it started drifting badly and
did a 'jump' of half a second at about 40 minutes.


Robert Jenkins.






It's more than just "not locking to the GPS". It thinks the GPS clock is
"insane"; that's the meaning of the "x" in the first column. The ntpq
display looks odd in a few other ways.

Are you actually using a server in Romania (.ro) and another in Mexico?
Why? Your servers should, ideally, be close to you in net space. Net
space is not the same as geographical space; distances in net space can be
longer but are seldom shorter than in geographical space. I'm having
difficulty thinking of a place that is close to both Mexico and Romania.

Are you setting your as closely as possible to the correct time before
starting ntpd? Or, better yet, starting ntpd with the -g switch so that
ntpd will set the clock on a one time basis?

It appears that you have set MINPOLL for server gate.jrw.intra to a value
of four. Why? The default values of MINPOLL and MAXPOLL are correct for
most situations.

The servers appear to have wildly differing opinions as to what time it
is. This is not good. If the servers don't have a clue, your clock
probably won't either. Add servers until you find at least four that show
something like close agreement.

I would look for servers closer to home, wherever that might be. Ideally
you should have values of delay less than 20 milliseconds. I would
suggest a minimum of four servers in addition to your GPS, at least until
you get things working.

Finally, do you have software that will monitor the Garmin GPS receiver
and report things like the number of satellites it sees, the signal
strength for each, the position in the sky for each, etc? Such software
can tell you many useful things and give you an idea of how well things
are working.

Hi Richard,

the main problem turned out to be not having timex.h linked to where the ntp
source files were expecing it.

The results now are:
ntpq -c peers
remote refid st t when poll reach delay offset
jitter
==============================================================================
*GPS_NMEA(0) .PPS. 1 l 15 64 377 0.000 42.053
0.018
-gate.jrw.intra 130.88.200.98 3 u 1 16 377 0.209 -1.937
0.039
+tigger.lentil.o 195.66.241.3 2 u 28 64 377 38.095 1.135
1.006
+mozart.musicbox 129.6.15.29 2 u 56 64 377 36.672 2.784
101.162
-sites.urchin.ea 80.253.108.112 3 u 8 64 377 22.293 0.692
1.225

ntptime
ntp_gettime() returns code 0 (OK)
time c8133c6c.3c2b0000 Mon, May 15 2006 18:44:12.235, (.235031),
maximum error 44020 us, estimated error 1 us
ntp_adjtime() returns code 0 (OK)
modes 0x0 (),
offset 51.000 us, frequency -0.312 ppm, interval 4 s,
maximum error 44020 us, estimated error 1 us,
status 0x107 (PLL,PPSFREQ,PPSTIME,PPSSIGNAL),
time constant 2, precision 1.000 us, tolerance 496 ppm,
pps frequency -0.312 ppm, stability 0.152 ppm, jitter 1.000 us,
intervals 2995, jitter exceeded 84, stability exceeded 16, errors 1989.

The servers were (other than the local one) just random pool entries.
I've already switched it to some uk.pool servers.

The other local machine (gate.jrw.intra) is running on good, fixed sources
with permission from the owners.
It's reasonably accurate so having a small maxpoll gets things locked rather
quicker, though this will probably not be needed if GPS/PPS setup proves OK.
This new setup will replace the other eventually, and I will then use the
servers presently configured on that.

The GPS aerial is in a window at present, if everything looks to be working
well I will change to a roof mounted one.

Regards,
Robert Jenkins.


.

Relevant Pages

  • Re: Best solution for millisecond accuracy
    ... I am running several servers on LAN with these restrictions: ... such as from GPS then you may have to go for an atomic clock onsite. ... clock like ntpd. ...
    (comp.protocols.time.ntp)
  • Re: Two time sources with offset => fails to synchronize
    ... It seems that the Radio clock has 50ms offset compared to the GPS ... time sources is the worst configuration you can have. ... have configured a local GPS clock and 3 external servers then the external ...
    (comp.protocols.time.ntp)
  • Re: Is 24PPM an Excessive Real-Time Clock Correction?
    ... I have no experience with other servers. ... The value of "drift" that gets settled on after a few days is 24 PPM. ... Consider that a computer manufacturer typically spends something like $2 US on the components for the clock! ...
    (comp.protocols.time.ntp)
  • Re: Is 24PPM an Excessive Real-Time Clock Correction?
    ... I have no experience with other servers. ... The value of "drift" that gets settled on after a few days is 24 PPM. ... Consider that a computer manufacturer typically spends something like $2 US on the components for the clock! ...
    (comp.protocols.time.ntp)
  • Re: Is 24PPM an Excessive Real-Time Clock Correction?
    ... I have no experience with other servers (to know how accurate the ... The value of "drift" that gets settled on after a few days is 24 PPM ... You say "the clock" like as if there were only one. ... other sources listed - most commonly the PIT or APIC timer AFAIK. ...
    (comp.protocols.time.ntp)