[ur-wg] Aggregate Accounting

[Matthew Ford]

Quoting Rosario Michael Piro <piro at to.infn.it>:

> I think a usage record should be explicitly declared to be aggregate 
> (how should be discussed), otherwise we would easily undermine all 
> standardization efforts. How does for example an arbitrary 
> implementation of the Resource Usage Service (RUS), that uses the UR, 
> realize that your UR is meant to be aggregate. It would most probably 
> (and usually should) interpret the StartTime as the start time of job 
> execution and the End Time as stop time of job execution of a single 
> job. It wouldn't understand urwg:description="accountperiod" as 
> modifier of the meaning of Start and EndTime.

So, my point here, is that section 10 of the spec defines StartTime, Endtime,
and CPUDuration as relating to Usage *not* job.  The descriptor is used to
attach meaning but it's not actually necessary, it's essentially a free
element.  Without the descriptor field the aggregate example is still valid.

I guess the issue is that the current section 10 defs allow this; I think
deliberately. For example the CPUDuration field can be used in a number of
circumstances: aggregate info is one of them, time on a batch system another,
cputime spent copying files may be yet another.

If we do move these fields into seperate sub-elements or specific "Job" types
(which I'm in favour off) then we have to be very careful.  Anyone using the
schema as I think original intended will have to made aware of these
significant changes.

I'd love to understand from the original authors if this is actually how they
envisaged its use.

> Of course you might customize the behaviour of your RUS (or whatever 
> service you use for accounting), but that would make it a 
> non-standard solution.
>
> But I agree that an aggregate format would be most useful. If it will 
> not be defined by the UR-WG then most probably most Grid 
> environments/projects will define their own aggregate format. This as 
> well would risk to undermine the standardization efforts.
>
> Before declaring an aggregate UR out of scope, we should at least 
> start (via the mailing list) a discussion on what would be necessary 
> (new elements? Or simply an additional attribute 
> 'urwg:scope="aggregate"' to some of the elements? To which? Allow 
> multiple job IDs for aggregate URs? etc.
>
> But I think this would be a longer process and wouldn't lead to quick 
> results. The improvement of the current UR for single jobs will (and 
> should) have the priority I guess.
>
> Cheers,
>
> Rosario.
>
>>
>> Matt.
>>
>
>
> -- 
> -------------------------------------
>    Rosario Piro (piro at to.infn.it)
>     http://www.to.infn.it/~piro/
> -------------------------------------
> Istituto Nazionale di Fisica Nucleare
> Sezione di Torino
> -------------------------------------
> National Insitute for Nuclear Physics
> Section of Turin, Italy
> -------------------------------------
>