Re: Objy/ROOT discussion at today's phone meeting

[Arie Shoshani <arie@math.lbl.gov>]

> For the long term, I propose:
> If we use ROOT files _only_, it is no problem for the Query 
> Estimator/index builder to make a mapping from file name to
> FID to be used internally by the Storage Manager.
> If there will be a mix of Objy and ROOT files, then the problem
> of uniqueness arises. I.e., QE has to choose FID's that aren't
> in use and won't become used by Objy. This could be solved by
> having the index builder scan the Objy catalog and generate 
> FID's that are not already taken, _each time_ new files 
> (ROOT or Objy) are added.

The FID name space is indeed a problem.  However, if we use 32 bit IDs 
(which will be the case when Objectivity goes to containers being 
files) we can reserve the high order bit to distinguish between 
Objectivity IDs and ROOT IDs.  But, this will require the cooperation 
of Objectivity with such a scheme.

 
> For MDC2 the index building process and the QE can handle
> that files for different components being in different
> directories on HPSS. This information is passed through
> the component.hpss.paths file in the index building
> process (the details of which are yet to be released, but
> I'll do that soon).

This will work for mapping into a single directory in Objectivity.
For the general solution we'll need an Objectivity to HPSS path 
mapping.

Arie.