Re: time out mechanism for QO/QE

["R. Jeff Porter" <rjporter@lbl.gov>]

I also like the swap-to-disk model.   I have been concerned about
the user problems incurred when the QO is used to send the "done"
message to the QE at QO-destruction. In this new model, rather than
having the QE expunge the query, the "done" message would signify
that this query's information is canidate for swapping to disk.  The
real "done" would then come via the OOI-QM route.

    Jeff

Douglas Olson wrote:

> I like the "swap to disk" model. I think the following scenario
> may turn out to be a popular one:
> 1. Physicist runs some GUI with QO in it, makes
>    a new query, possibly even does execute on it.
>    Gets query token, then exits GUI.
> 2. Some time later physicists submits batch jobs
>    using this query token.  Parallelism is acheived
>    by multiple batch jobs with single query token.
>    When some iterator gets the message that the final
>    event in the query has been delivered, it should
>    issue the done message back to QM.
>
> There could be a long time between 1 and 2, with no
> active QO in existance but the query is still valid.
> It would be useful for the persistent query to have
> info like owner, date/time created  that an administrator
> could use to clean up "dead" queries, but I don't think
> there should be any automatic timeout.  It could be
> implemented that the QO could specify a lifetime for
> a specific query when it is created, but that seems
> like an unnecessary refinement for now.
>
> Doug
>
> > -----Original Message-----
> > From: Henrik Nordberg [mailto:hnordberg@lbl.gov]
> > Sent: Wednesday, November 11, 1998 2:15 PM
> > To: gcdev
> > Subject: time out mechanism for QO/QE
> >
> >
> > Hello,
> >
> > since QE is keeping information in memory for each query, it's desirable
> > to get rid of "dead" queries. How about having the QO ping the QE
> > every now and then. If a query has been idle for a certain time, then
> > QE could delete it and abort() it. An alternative would be to swap it
> > to disk.
> >
> > ?
> >
> >  - Henrik
> > _________________________________________
> > Henrik Nordberg       <hnordberg@lbl.gov>
> > Scientific Data Management Research Group
> > Lawrence Berkeley National Laboratory
> >