[LinuxPPS] a fluke...?

[James Boddington]

gnu not unix wrote:
> http://www.wraith.sf.ca.us/ntp/rrd/index.html
> 
> 
> Is anyone else running  plots and seeing any interesting
> trends? 

Some static graphs from earlier today
http://www.users.on.net/~boddingt/ntp/rrd/

ntp is my ntp server
barney gets its time from ntp.

The time server is a dell gx1, p2-350 with a garmin gps25-hsv. Pps via the 
serial port. Linux kernel is 2.6.20-rc5. The '2 days' offset graph for ntp is a 
good example of what I have been finding with kernel >= 2.6.18.

Before Thur 12:00 on that graph I had minpoll 4 & maxpoll 10. It was generally 
staying within +/- 30us. At 12:00 I changed minpoll to the default of 6 and now 
get +/- 150us.

I get far better performance with 2.6.17. Very similar to the current +/- 10us 
plots on your site for wraith. Freebsd or 2.4.33.3-NANO better again.

You mentioned spikes in a previous message to this list. I have wondered what 
your loopstats looked like for the same period. The spike on my graph is not 
all it seems. The graph shows a spike of nearly 900us. The spike was actually 
-48ms and recovered very quickly with a small amount of overshoot. It was not 
picked up properly when sampling for rrd. Sampling is done every 5 minutes.

I get a few spikes with linuxpps. I eventually wrote a small daemon to take a 
sample of pps readings, drop the outliers, average the rest then use the SHM 
driver to pass the time stamp onto ntpd.

-- 
    James