Hi Tina, We are running a test sensitivity analysis on caparmor with the new version of ISIS-Fish. With the new version we use the ram instead of the disk when possible, so accesses to the disk should have been reduced drastically. It seems to be running OK, and files do not appear on the disk so everything seems to be working as planned, but simulation time did not really decrease. Could you check whether we use the disk and the ram properly (if I understood well we should be using much less disk space and a bit more ram but not much as results are erased while the simulation is running), so we can understand if simulation time still comes from writing things on the disk or if it is due to something else ? Loïc Le 13/02/2014 17:42, Tina ODAKA a écrit :
hi stephanie, as i told you on phone,
if they use option -q isisfish with account of loic, it use max memory 2.9G and 3 hours of wall time.
I'll check tomorrow afternoon, how it is running. and if it runs ok, i try to put more core during weekend.
regards, Tina
Le 13/02/2014 15:39, Loic.Gasche@ifremer.fr a écrit :
Hi Tina,
Did you get the e-mail about writing the results.mapped file created by ISIS in the ram ?
It seems that it is not possible to do it in the short-term as it implies making some deep changes to the way ISIS works.
Do you have any idea about when enough cores would be free on caparmor for us to submit our sensitivity analysis ? In fact we will need to run two of these to test for contrasted management scenarios, each corresponding to several thousand simulations.
Loïc
lgasche@ifremer.fr a écrit :
Hi Tina,
I cannot submit it today but I will do it tomorrow.
Tina ODAKA <Tina.Odaka@ifremer.fr> a écrit :
Hi, i want to see how your job runs well on caparmor. tell me when you submit your simulation, i'll check how it runs. (if today after 16:30, if tomorrow i can not chek morning, it have to be afternoon)
regards, tina
Le 06/02/2014 11:31, Loic GASCHE a écrit :
Hi Tina,
It seems that we have solved the last few problems we had with ISIS-Fish and that it now runs well on caparmor.
We would like to run one big sensitivity analysis on caparmor.
So far we ran 500 simulations. The ideal for us would be 10000 simulations to be consistent with a sensitivity analysis we ran at the beginning of 2013. If this number is too high maybe we can make choices and reduce it a bit but we will still need to run several thousand simulations.
We are not in a great hurry so it is perfectly fine for us if the analysis runs for several weeks or even a month.
Is it possible to run the analysis on caparmor with "a few" cores and "automatically" increase the number of cores used when more cores are available on caparmor (weekends, holidays ?) ?
Do you think we should wait for a period when few jobs run on caparmor and only then run our big sensitivity analysis ?
Loïc
-- =================================================== Tina Odaka RIC - IDM - IMN - IFREMER Pôle de Calcul Intensif pour la Mer (PCIM)
Tel: +33 (0)2 98 22 41 85 Fax: +33 (0)2 98 22 45 46 email: Tina.Odaka@ifremer.fr http://www.ifremer.fr/pcim ==================================================