Overwriting a distributed job name by purpose

[Svenn Are Bjerkem]

Hi,

working distributed in a grid environment is quite fun. Problem is only
that the job name always has to be unique. This cause a bit problem for
me when I want to rerun a particular simulation. I usually give my
simulation jobs feasible names and not just "job". If I try to reuse a
name then the system complains that it is not unique. This is not that
much a problem when running interactive. The problem comes when I want
to automate the task of making plots by OCEAN. The job is divided
between two OCEAN scripts: one that runs the simulations and one that
performs the evaluation.

The reason is that I want to be able to rerun the simulations on an
interactive basis, but the evaluation is done in one batch as soon as
all simulations are in a state where I can say that the data are stable.
I have some 20 different simulations on the same top level (but with
different configurations down the hierarcy) and I want to compare all
these 20 simulations the way I do a corner simulation. (overlay). Here
the naming problems occure: If I have to rerun one of these 20
simulations manually, the name change and I have to change the
evaluation OCEAN script also.

Now I hope you understand why, and then the question is how to *force*
ADE to accept an already existing job name and overwrite those existing
data.

--
Svenn

 

[Satya Mishra]

"Svenn" == Svenn Are Bjerkem <svenn.are@bjerkem.de> writes:
Svenn> Date: Wed, 18 May 2005 12:40:47 +0200

Svenn> Hi,

Svenn> working distributed in a grid environment is quite
Svenn> fun. Problem is only that the job name always has to be
Svenn> unique. This cause a bit problem for me when I want to
<snip>
Svenn> Now I hope you understand why, and then the question is how
Svenn> to *force* ADE to accept an already existing job name and
Svenn> overwrite those existing data.

Svenn

I have used the shell script at finish option in the job submit form
to achieve this end. It's not elegant and fails rather frequently, but
was good enough for my purposes. As it is, it will select the most
recent simulation job in ADE. There is some code commented out that
will delete old jobs.

As usual, use at your own risk.

Satya

 

[Andrew Beckett]

On Wed, 18 May 2005 12:40:47 +0200, Svenn Are Bjerkem <svenn.are@bjerkem.de>
wrote:

Hi,

working distributed in a grid environment is quite fun. Problem is only
that the job name always has to be unique. This cause a bit problem for
me when I want to rerun a particular simulation. I usually give my
simulation jobs feasible names and not just "job". If I try to reuse a
name then the system complains that it is not unique. This is not that
much a problem when running interactive. The problem comes when I want
to automate the task of making plots by OCEAN. The job is divided
between two OCEAN scripts: one that runs the simulations and one that
performs the evaluation.

The reason is that I want to be able to rerun the simulations on an
interactive basis, but the evaluation is done in one batch as soon as
all simulations are in a state where I can say that the data are stable.
I have some 20 different simulations on the same top level (but with
different configurations down the hierarcy) and I want to compare all
these 20 simulations the way I do a corner simulation. (overlay). Here
the naming problems occure: If I have to rerun one of these 20
simulations manually, the name change and I have to change the
evaluation OCEAN script also.

Now I hope you understand why, and then the question is how to *force*
ADE to accept an already existing job name and overwrite those existing
data.

Svenn,

OCEAN has a function for this:

deleteJob(jobName)

So if you do:

name1=run(?jobName "myjob")
name2=run(?jobName "myjob")

name1 will be myjob, whereas name2 will be myjob001.
If I then do:

deleteJob(name1)
name1=run(?jobName "myjob")

name1 will be myjob again.

Regards,

Andrew.