John, The only contentious difference between what you are arguing for and what I am arguing for is that in your case you require the grid to perform the interpolation. I argue that the grid, as one relying party, is _not_ the best place for this, as URs may go anywhere. If a document bearing only the state of the system at an instant in time does not go into the system with the logic to interpret that data (i.e. the algorithm and the sampling rates) then those documents are worthless. Therefore, I hold to my opinion that any interpolation must be done at the UR creation point (i.e. in the domain of the resource provider) regardless of whether their storage system has journalling, relies upon sampling or employs whatever alternative means to arrive at an integrated value. In this light I do not believe your argument has any weight stating that it is too difficult to supply an integral value when, at a bear minimum, the resource provider can perform locally the same algorithm that you would have the accounting system perform remotely. Further: Assuming sampling, the storage system has better indicators of how its facility is used. It will therefore be in a better place to increase sampling rates and adjust the algorithm used to achieve more accurate usage values should it need to do so.
Mike, I think you are way out of line here. As long ago as Munich OGF we discussed time integrals for UR and I thought we reached a consensus (well maybe everyone except you) that we could not ask storage systems to start evaluating integrals of storage usage for the use of UR. They don't do it and we would be asking them to implement journaling or to recalculate usage at every file interaction (or at least write/modify/delete). Just not on.
I do not believe I am "way out of line". I am not the only person to have the view that integrated values are a requirement of a definition of usage. The following people were at the Munich meeting: Gordon, John - STFC Kennedy, John - Rechenzentrum Garching (RZG) Keskrand, Kalle - EENet Kretzschmar, Michael - MNM-Team Maier, Andreas - LRZ Garching Raitviir, Tõnu - EENet Reetz, Johannes - Max-Planck-Institut für Plasmaphysik URBAH, Etienne - LAL, Univ Paris-Sud, IN2P3/CNRS Wolfrat, Jules - SARA I believe Etienne also shares my view strongly about integrals vs instantaneous values. I was not able to attend UR-WG in Munich as it clashed with security sessions. I am sorry for this. I believed you had a reasonable understanding of what would constitute a storage usage record. At that time I was happy to leave it at that.
I thought we all understood that this is a compromise for the pragmatic purpose of getting a UR soon that can be implemented in the real world.
I cannot find minutes of that meeting only the preprepared slides. The slides do not (in fairness as they were probably written prior to discussion cannot) support that statement of compromise.
You seem to enjoy the tautological discussion of what accounting should be in the theoretical world. I just want to start writing records tomorrow, if not today.
Please leave the rhetorical aspersions out. My "Tautological discussion" as you put it is not so. I repeat myself merely to put across the same argument in different words so as to help you and the group understand where I'm coming from. -- I have not yet heard a counter argument strong enough to make me put-up and shut-up.
In the longer term maybe we make requirements against storage systems that they implement integrals as well as user level and i/o accounting.
For now can we concentrate on agreeing something that can be implemented.
Any of this can be implemented and easily. And as I understand UR v1 should be sufficient for your needs as it stands. Mike