RE: file position on tape

["Douglas Olson" <dlolson@lbl.gov>]

I am glad to hear that HPSS will optimize on pending pftp requests.
I wondered that even if we tried to request files in an optimum
tape order that there still maybe some work to get it acting right.
I know that tape position is not on the MDC2 list so I am perfectly happy
to take it up after MDC2.  In the long run, it may be that
file order on tape may be one of the more effective optimizations.

Doug

> -----Original Message-----
> From: Arie Shoshani [mailto:arie@math.lbl.gov]
> Sent: Wednesday, February 03, 1999 1:34 PM
> To: gcdev@sseos.lbl.gov; dlolson@lbl.gov
> Subject: Re: file position on tape
>
>
>
> > Dave Z. and I spoke with John Riordan (and Ed and Razvan) yesterday
> > about getting information of where files are on tape.
> > The thing that looks like it should be easy to do is to
> > get the list, for a class of service (COS), of
> > filenames, tape number, and position (order) of file on tape.
> > I was thinking that you could use this along with the
> > list of files that go into the GC index to include the
> > information about tape number and order on tape.
> > Then it would be possible (conceptually) to include this
> > tape information in the policy decisions about what
> > order to retrieve files.
> > Dave will be back at LBL on Thursday and can talk to you
> > about it, and probably show you an example of a file that
> > has the tape information.  If you recall, he got information
> > like that about the files we used at the end of MDC1.
> > Why don't you think about that and consider whether it
> > is possible to use it in MDC2.
> >
> > Thanks,
> > Doug
>
> Doug,
>
> That can be very useful information, so we should try to make it available
> in the index.  But, as Alex said any policies that use tape information
> should be left to be done after MDC2.
>
> BTW, taking advantage of tape information is partially done by HPSS if we
> request sufficient number of PFTPs.  If many PFTPs are in the
> HPSS queue it
> optimizes the order of reading them.
>
> Arie.
>
>