UWS PENDING Destruction Time can be immediate

[Pierre Le Sidaner]

Hi Petr

The description of hand made job scheduler should be done by a common 
batch queue scheduler, try : Slurm, Torque Maui, sun grid engine ... 
Then you will win a lot of time. This softs are made to optimise what is 
going to run on the machine, and the choose of what to run will be done 
automatically, on the criteria you have configure.
This softs have been optimise by system engineer since some time and 
they used it also in large cluster. UWS is made to launch job, not to 
take place of the job scheduler.

Regards
Pierre

On 11/12/2014 10:07 AM, Petr Skoda wrote:
> Hi all
>
> just briefly looking in ongoing discussion I am wondering why someone 
> has again the tendency removing something important from UWS - 2 years 
> ago it was ABORT / DELETE discussion.
>
> In my opinion the UWS has already proved its clever design in many 
> cases. Maybe you remember my special use case called VO-CLOUD which 
> contains interactive experiments (e.g. data mining of spectra 
> profiles) on master which uses UWS to launch the real computation on 
> some "workers " using UWS.
>
> We perfectly exploit the PENDING when the master server get uploads of 
> all (big) input files and parameters and selects the posible workers 
> (taking into account their particular capability - i.e. the libraries 
> and binaries installed) and is in principal ready to lauch job when 
> asked to do so (it contacts the worker and tells him handle where he 
> can download the encrypted data for run).
>
> The trick is the user can upload (prepare) many experiments - so the 
> jobs are created (put into database of experiments etc ) which are in 
> PENDING state  and later (e.g. from a mobile phone) can decide which 
> one is to be run .....
>
> But from the we client of master server with experiments he can 
> immediately select from menu the "RUN immediately" which creates job 
> and sends it directly to worker using POSTIng PHASE=RUN ...
>
> So it is fully under user control what he wants to run later .
> The idea si to set a work for "whole week" and control from mobile the 
> status (checking images , logs outputs of convergence etc ....) he may 
> even decide to ABORT the job in PENDING phase when he sees the 
> nonsense of given parameters in an experiment (e.g. similar set did 
> not converge....)
>
> So we need both PENDING , immediate RUN, ABORT and DELETE
> in fact all what is in UWS...
>
> details may be seen in
>
>
> http://wiki.ivoa.net/internal/IVOA/InterOpOct2014Applications/defbanff-vocloud.pdf 
>
>
>
> Regards,
>
>
> Petr
> *************************************************************************
> *  Petr Skoda                         Phone : +420-323-649201, ext. 361 *
> *  Stellar Department +420-323-620361           *
> *  Astronomical Institute AS CR       Fax   : +420-323-620250           *
> *  251 65 Ondrejov                    e-mail: skoda at sunstel.asu.cas.cz  *
> *  Czech Republic                                                       *
> *************************************************************************


-- 
-------------------------------------------------------------------------
                            Pierre Le Sidaner
                         Observatoire de Paris

Division Informatique de l'Observatoire
Observatoire Virtuel 01 40 51 20 89
61, avenue de l'Observatoire 75014 Paris

mailto:pierre.lesidaner at obspm.fr
http://vo.obspm.fr

--------------------------------------------------------------------------