On 2016-10-03 17:03, Alessandra Forti wrote:
> Most UK sites are ok, there are however 3 problematic sites which have
> RED more often than not UCL, RHUL and ECDF. Brunel was red in August
> and September too.
I hope there is there universal agreement over whether the start of the
record or the end of the record signifies the month the record relates
to. This is a trivial factor, but experience shows that many accounting
systems errors fall into corner case traps.
And are operators that define periodicity equivalent to (A >= B and A <
C), where A is the epoch of the record, B is the starting epoch of the
period and C is the terminating epoch, e.g. B + D, where D is duration?
Again it's a trivial element, but many "plus or minus one errors" occur
at that boundary as well.
But these edge issues would tend to only account for small discrepancies
over a long time span. Of course, there is also the bench-marking issue,
but I hope covered most of those pitfalls in the GridPP talk:
https://indico.cern.ch/event/556609/contributions/2256481/attachments/1329531/1997362/benchAndPub.pdf
If the bench-marking and publishing are right, and the edge cases are
right, then by elimination I'd expect any errors to come from failure to
transmit the records (or receive them) or disconnects in the subsequent
storage and processing (where else could the errors arise?)
Cheers,
Ste
|