[ntp:questions] Is 24PPM an Excessive Real-Time Clock Correction?

Spoon root at localhost.invalid
Tue Jul 10 09:19:47 UTC 2007


Richard B. Gilbert wrote:
> Per Hedeland wrote:
>> Richard B. Gilbert wrote:
>>
>>> Spoon wrote:
>>>
>>>> Richard B. Gilbert wrote:
>>>>
>>>>
>>>>> David T. Ashley wrote:
>>>>>
>>>>>
>>>>>> I have ntpd running on a RHEL Linux Dell 1U rack server in another 
>>>>>> city.
>>>>>>
>>>>>> I have no experience with other servers (to know how accurate the 
>>>>>> clocks are or are not).
>>>>>>
>>>>>> The value of "drift" that gets settled on after a few days is 24 
>>>>>> PPM (about 14 seconds/week).
>>>>>>
>>>>>> This strikes me as a little high, because even my $20 Timex watch 
>>>>>> does better than this.
>>>>>>
>>>>>> Any insight into whether 24 PPM is excessive for my server?
>>>>>
>>>>>
>>>>> Consider that a computer manufacturer typically spends something 
>>>>> like $2 US on the components for the clock!
>>>>
>>>>
>>>> You say "the clock" as if there were only one.
>>>
>>> <snip>
>>> I meant the time of day clock; it's the only one that ntpd concerns 
>>> itself with.  I thought it would be obvious from the context!
>>
>>
>> The "time of day clock" is normally used to refer to the CMOS or
>> equivalent one that ticks even when power is off - that would be "RTC"
>> in the previous post I believe - it's the one clock that ntpd absolutely
>> does *not* concern itself with.
>>
>> The one it *does* concern itself with is the "system" clock, a software
>> clock that is advanced by the kernel based on interrupts from one of the
>> other sources listed - most commonly the PIT or APIC timer AFAIK. But of
>> course the interesting part isn't the timers (i.e. counters), but the
>> crystal(s) etc that drive them.
> 
> Sorry about that.  Too many clocks!  In any event, the crystal 
> oscillator gets very little attention from the manufacturer; the 
> components are just about the cheapest available and no attempt is made 
> to tune the crystal oscillator to bring it exactly on frequency.  It's 
> generally within 100 PPM of the desired value and that's plenty good 
> enough for most people.  If the customers cared, manufacturers would 
> include a higher quality crystal in a temperature compensated oscillator 
> circuit and adjust it at the factory to the correct frequency.
> 
> Most people buy computers by the G Hz of the processor, not the accuracy 
> of the clock.  Most don't really need the accuracy we can get with NTP. 
>  And a lot of people use NTP only to synchronize clocks to each other; 
> they couldn't care less what time it is as long as all the computers 
> agree on a time.

I think you missed my point.

The HPET is supposed to replace all the other clocks I mentioned.

Some manufacturers do choose high-quality components when they implement 
the HPET.

Intel south bridges have included an HPET for some time now. Perhaps 
other manufacturers will follow their lead?

We agree that the situation right now is pathetic. I was saying that 
maybe, just maybe, there is hope on the horizon?




More information about the questions mailing list