Re: time drift in camonitor timestamps

[Andrew Johnson <anj@aps.anl.gov>]

Hi Inari,

On 2013-01-08 inari badillo wrote:
> The magnitude of the error is about 700 microsecond (average), no matter
> the SCAN value is. So, it takes about 24min to have a 1s error.
> Answering Michael Davidsaver´s question, I mean precessing relative to the
> start time. The different does not increase every cycle. Sorry if it wasn´t
> clear.
> And finally yes, it seems reasonable to be caused by processing time. But
> in this case, wouldn´t it lead to malfunction?

The scan code does compensate for the time it takes to process the records in 
its chain, but it doesn't try to correct for the time difference between when 
the thread asked to be woken up and when the OS actually woke it up, and I 
think that is the source of your 700us delay.

After adding some code to measure that delay on my workstation (RHEL 6.3 on a 
Dual-Core 1GHz AMD Opteron) I have seen varying delays between -60us and 
+230us, but nothing quite as big as 700us; my average is around 30-50us.

I will look at merging Eric's patch, or something like it (thanks Eric!).

- Andrew
-- 
There is no such thing as a free lunch.  When invited for lunch,
it is best to check if you are there to eat, or to be eaten.
-- Clive Robinson