From wolbers@fnal.gov Mon Mar  8 17:52:58 1999
Date: Sun, 29 Nov 1998 09:49:46 -0600
From: Stephen Wolbers <wolbers@fnal.gov>
To: chenyc@fnal.gov
Subject: [Fwd: About floating worker assignments]

Yen-Chu,

    Maybe we can discuss this during the video-conference.  I think Heidi's
suggestion was just for the prototype farm, which is not such a bad idea.  I
agree with Ping that sharing nodes on the "real" farm is not a good idea but on
the prototype I think it is worth a try.

Steve
    [ Part 2: "Included Message" ]

Date: Sun, 29 Nov 1998 07:48:16 -0800 (PST)
From: Ping Yeh / Academia Sinica <pyeh@phys.sinica.edu.tw>
To: Heidi Schellman <schellma@fsgi02.FNAL.GOV>
Cc: Marilyn Schweitzer <marilyn@bastet.fnal.gov>, schellma@fsgi02.FNAL.GOV,
     run2farms@fnal.gov
Subject: About floating worker assignments

Hello Heidi,

    I can imagine that theoretically floating working assignments
is more efficient in CPU utilization.

    However, practically, it may produce problems.

    Under what condition should a node be granted from one experiment
to another?  CPU utilization?  If so, under what condition should the
node be granted back to original experiment when they need it?

    I feel that this will produce some kind of "political" problems.

							Regards,
							Ping

On Fri, 27 Nov 1998, Heidi Schellman wrote:

> Would  CDF and D0 be willing to have floating worker assignments via queues?(
> be able to run on any machine, not just a few assigned machines but never have 
> a CDF and D0 queue on the same machine at the same time?)
> 
> That would allow us to dynamically have more (or less) workers depending on
> conditions, with farms group personnel reassigning queues to machines.
> 
> 
> I think the D0 disk layout allows this, how about CDF?
> 
>  Thanks, Heidi
> 
> 
> 
>