[DRMAA-WG] Torque/PBS DRMAA - drmaa_wi* POSIX equivalents II

[Krzysztof Kurowski]

Hi Andre, All,

Andre Merzky pisze:
> Hi Andreas, 
>
> Quoting [Andreas.Haas at Sun.COM] (May 08 2007):
>   
>> Actually the idea of having kind of a general pupose state definition 
>> framework model was discussed very, very early within DRMAA-WG. At that 
>> point in time the workgroup decided not to pursue this and I think this 
>> was good since I never heard such a request from the DRMAA user community.
>>     
>
> Well, DRMAA was around much earlier than BES or SAGA or many
> other groups - so I'm certainly not saying you should have
> used this model :-)
>
> Also, you are right: don't fix it if it ain't broken...  If
> users are happy, than thats a good measurement for
> success...
>
> Cheers, Andre.
>   
I do support Andre's comments... I think a state definition model in 
next DRMAA specs
should be improved or at least discussed again to reflect some new 
requirements
and also best practices and ideas taken from SAGA and BES.

I'd like to encourage you to check out the section 3.1.2. Job States we 
proposed
in Open DRMAA Service Provider (OpenDSP) v1.0
http://downloads.sourceforge.net/opendsp/dsp-manual-1.0.tar.gz?modtime=1162140031&big_mirror=0

Actually, when we have been discussing next OpenDSP releases a new 
question appeared:
how/when can we support BES 1.0? Hence, we are interested in the 
convergence among DRMAA, SAGA
and BES discussions around state model as well as new routines for job 
status change monitoring
such as transition into the PENDING and RUNNING states.

Thanks,
Chris

>   
>> Regards,
>> Andreas
>>
>> Until now DRMAA users have not stated demand.
>>
>>
>> On Sat, 5 May 2007, Andre Merzky wrote:
>>
>>     
>>> The nice thing about the BES (and SAGA) state model is that
>>> it is rather simple, and probably represents a common
>>> denominator for many systems.
>>>
>>> At the same time, it can easily cover backend specific
>>> substates (e.g. the DRMAA distinction for two different
>>> SUSPENDED states would be modeled as substates to one
>>> SUSPENDED state).  These substates are, however, (mostly)
>>> for informational purposes.
>>>
>>> Anyway, I'd like to encourage you to check the BES/SAGA
>>> state model in detail.  Its probably better described in the
>>> BES doc, but possibly easier to understand in the SAGA spec
>>> ;-)
>>>
>>> It would be nice if we could agree on one common job state
>>> model at some point in the future...
>>>
>>> Cheers, Andre.
>>>
>>>
>>> Quoting [Peter Troeger] (May 04 2007):
>>>       
>>>> From: Peter Troeger <peter.troeger at hpi.uni-potsdam.de>
>>>> To: DRMAA Working Group <drmaa-wg at ogf.org>
>>>> Subject: Re: [DRMAA-WG] Torque/PBS DRMAA - drmaa_wi* POSIX equivalents II
>>>>
>>>>         
>>>>> Issue:
>>>>>
>>>>> Removing drmaa_wif* routines and introducing one-call semantics could
>>>>> lead to more clear, consistent and informative job life cycle
>>>>> graph. For
>>>>> example, additionally to the existing job states returned by
>>>>> drmaa_job_ps(), we could imagine also states such as
>>>>> DRMAA_JOB_PS_COREDUMPED or DRMAA_JOB_PS_SIGNALLED.
>>>>> ---------
>>>>>
>>>>> If the drama_w* functions are present, this is not necessary.
>>>>>
>>>>> If one day the drama_wif* routines are gone, we might need extensions
>>>>> mechanism (like some other groups, JSDL or SAGA?) for the job
>>>>> states to
>>>>> avoid explosions of states and implementation complexity.  This is
>>>>> clearly future versions staff.
>>>>>           
>>>> OGSA-BES had some long discussion about job states, we should look on
>>>> this. JSDL is no option here, since their job is over after the job
>>>> submission.
>>>>
>>>>  I vote for an updated field study, where we compare the available
>>>> job state information in the most prominent DRM systems. The key is
>>>> to find the least common denominator - therefore this would be the
>>>> chance to remove the different SUSPENDED states ;-)
>>>>
>>>> Peter.
>>>>         
>>> -- 
>>> "XML is like violence: if it does not help, use more."
>>>
>>> --
>>> drmaa-wg mailing list
>>> drmaa-wg at ogf.org
>>> http://www.ogf.org/mailman/listinfo/drmaa-wg
>>>
>>>       
>> http://gridengine.info/
>>
>> Sitz der Gesellschaft: Sun Microsystems GmbH, Sonnenallee 1, D-85551 
>> Kirchheim-Heimstetten
>> Amtsgericht Muenchen: HRB 161028
>> Geschaeftsfuehrer: Marcel Schneider, Wolfgang Engels, Dr. Roland Boemer
>> Vorsitzender des Aufsichtsrates: Martin Haering
>>     
>
>
>
>