[ntp:questions] 500ppm - is it too small?

David Lord snews at lordynet.org
Fri May 8 04:39:46 UTC 2009


Richard B. Gilbert wrote:
> E-Mail Sent to this address will be added to the BlackLists wrote:
>> Richard B. Gilbert wrote:
>>> The specifications for NTPD say that it will correct
>>>  errors less than or equal to 500 PPM.
>>>
>>> I beleive that hardware outside of this limit can
>>>  properly be described as broken!
>>
>> Could this be mitigated by setting panic to 0?
>>
>>  (Wasn't there some other bigtimestep option too?)
>>
>>
> 
> You'd have to ask Dave Mills.  I'd just scrap out the mother board or 
> the whole computer!  If it's software that's causing your pain it may be 
> unreasonable to ask you to scrap out Windows.
> 
> It's not unreasonable to expect a computer clock to have a frequency 
> error of less than 50 Parts Per Million and NTPD will cope with up to 
> 500 PPM.  Remember that 500 PPM works out to 43 seconds per day!  As I 
> said: "SERIOUSLY BROKEN"!!

On various NetBSD systems I have an adjustment in the kernel
configuration which can bring the system clock within 500ppm.

# c7x1200c_02
# c7x1200c approx -48ppm so increase TIMER_FREQ +48
# added:
# options TIMER_FREQ=1193494
#
# c7x1200c_01
# c7x1200c approx -268ppm so increase TIMER_FREQ +264
# added:
# options TIMER_FREQ=1193446
########################
# default TIMER_FREQ=1193182

Another motherboard of same make model for NetBSD had
adjustment total +292.

Above was needed on NetBSD 3.1 but there is a note that on
4.0 same motherboards didn't need any adjustment so I guess
were within 30ppm. I can't locate dmesg for 4.0 to confirm
this but suspect CPU misidentified on 3.1.

I've used same to correct for near 1000ppm error and ntp
performance has then been no better or worse than other systems.

I don't believe it's the system clock crystal that's out, rather
it's how the OS sees the hardware.


David




More information about the questions mailing list