[Pgi-wg] OGF PGI : Using GLUE 2.0 inside JSDL - Requirement 100 (JD3)
Etienne URBAH
urbah at lal.in2p3.fr
Thu Dec 9 12:40:13 CST 2010
Morris, Johannes and all,
Concerning OGF PGI requirement 100 :
http://forge.gridforum.org/sf/wiki/do/viewPage/projects.pgi-wg/wiki/ReqJD3
The Job Description document MUST reference all grid entities in
conformance to the GLUE specification
From the raw minutes and my own memory of today's telephone conference,
I have extracted the points directly concerning requirement 100
(excluding points concerning the PGI process).
I have written down these points in the above mentioned wiki page, and
here :
* Mark: What does 'values do not match' mean ?
* Etienne: This means that the values describe the same architecture
with different names.
* Oxana: This means that the enumerations are not the same (bottom line).
* Andrew: I agree to align to GLUE2, to use space additions for
enumerations when 2 groups call them differently.
* Morris: We need a general strategy to integrate GLUE2 details inside
JSDL.
* Agreement on 'Example of GLUE2 extension and integration inside JSDL'
http://forge.gridforum.org/sf/docman/do/downloadDocument/projects.pgi-wg/docman.root.input_documents/doc16166/1
:
o We integrate GLUE2 elements in some parts of the JSDL job
description
o JSDL should be further used and not replaced, but augmented
with GLUE2.
* Morris: What is your feeling of re-using the GLUE2 parts that are
compatible ?
* Etienne: Some GLUE2 attributes do NOT conflict with JSDL, and can be
directly integrated inside JSDL.
As first candidates for integration inside JSDL, I propose
the GLUE2 attributes of the 'ExecutionEnvironment' class (Chapter 6.6 of
GLUE 2.0).
* Morris: We have to prepare a discussion on these candidates for the
next telephone conference.
* Etienne: I will contribute to this preparation.
* Morris: Are the attributes of the 'ComputingService' class of GLUE2
also good candidates for integration inside JSDL ?
* Etienne: These attributes are only numbers of jobs, therefore I think
that they are NOT good candidates. I prefer to focus on
'ExecutionEnvironment' attributes, which describe hardware.
Do not hesitate to edit the wiki page if points on this precise
requirement have been missed or inaccurately written down.
Do not hesitate to add your own comments and suggestions for improvement
at the end of this wiki page.
In particular, after more careful thinking, many attributes of the
'ComputingService' class of GLUE2 are good candidates for integration
inside JSDL indeed.
Clients may for example issue requirements on computing services having :
- given values for 'Capability',
- a given 'Type' or 'QualityLevel',
- low values for 'WaitingJobs', 'SuspendedJobs', ...
Anyway, I will focus first on 'ExecutionEnvironment' attributes, and
then on attributes of other classes.
Best regards.
-----------------------------------------------------
Etienne URBAH LAL, Univ Paris-Sud, IN2P3/CNRS
Bat 200 91898 ORSAY France
Tel: +33 1 64 46 84 87 Skype: etienne.urbah
Mob: +33 6 22 30 53 27 mailto:urbah at lal.in2p3.fr
-----------------------------------------------------
On Wed, 08/12/2010 18:46, Etienne URBAH wrote:
> Morris, Johannes and all,
>
> Concerning the usage of GLUE 2.0 inside JSDL :
>
> ...
>
> +------------------------------------+
> | Detailed work on requirement 100 |
> +------------------------------------+
> 100 :
> http://forge.gridforum.org/sf/wiki/do/viewPage/projects.pgi-wg/wiki/ReqJD3
> The Job Description document MUST reference all grid entities in
> conformance to the GLUE specification
>
> Rationale
> ---------
> Figure 1 of the 'Job Submission Description Language (JSDL)
> Specification, Version 1.0' at http://www.ogf.org/documents/GFD.136.pdf]
> clearly shows that, in order to work correctly, a 'Superscheduler, or
> Broker, or …' has to match client requirements expressed in the JSDL
> document to information provided by a Grid Information System.
>
> Therefore, it is essential that client requirements on entities
> described by GLUE use only GLUE classes and attributes.
>
> For XML purposes, I propose to use the 'glue' namespace specified at
> http://schemas.ggf.org/glue/2008/05/spec_2.0_d42_r01
>
> JSDL elements clearly incompatible with GLUE 2.0
> ------------------------------------------------
> Searching for JSDL elements clearly incompatible with GLUE 2.0 at
> http://www.ogf.org/documents/GFD.147.pdf I have found following ones
> (but careful study by other persons could perhaps find other ones) :
>
> - JSDL 'ProcessorArchitectureEnumeration' specified in JSDL section
> 5.2.1 and used by the 'CPUArchitectureName' element specified in JSDL
> section 6.4.15 :
> The corresponding GLUE 2.0 enumeration is 'Platform_t' specified in GLUE
> 2.0 appendix B.26.
> While some JSDL values (such as 'powerpc' and 'sparc') match GLUE 2.0
> values, some other JSDL values (such as 'x86_64', 'x86_32' and 'ia64')
> do NOT match GLUE 2.0 values (which are respectively 'amd64', 'i386' and
> 'itanium').
> Therefore, I propose to deprecate JSDL
> 'ProcessorArchitectureEnumeration' and replace it with GLUE 2.0
> 'Platform_t' enumeration (for XML, inside the above described 'glue'
> namespace).
>
> - JSDL 'OperatingSystemTypeEnumeration' specified in JSDL section 5.2.3
> and used by the 'OperatingSystemName' element specified in JSDL section
> 6.4.12 :\\
> The corresponding GLUE 2.0 enumerations are 'OSFamily_t' specified in
> GLUE 2.0 appendix B.23 and 'OSName_t' specified in GLUE 2.0 appendix
> B.24.\\
> GLUE 2.0 permits the client to specify the desired OS either through a
> vague requirement ('OSFamily' attribute) or through a precise
> requirement ('OSName' attribute).\\
> Therefore, I propose to deprecate JSDL 'OperatingSystemTypeEnumeration'
> and replace it with the GLUE 2.0 'OSFamily_t' and 'OSName_t'
> enumerations (for XML, inside the above described 'glue' namespace).
>
> Practical usage of the 2 previous proposals permitting backward
> compatibility should be easily achieved using the JSDL 'other' value as
> described in JSDL section 4.3 'Semantics of the JSDL “other” Value'.
>
>
> Please study and criticize the above proposals. Do not hesitate to add
> comments and suggestions for improvements at
> http://forge.gridforum.org/sf/wiki/do/viewPage/projects.pgi-wg/wiki/ReqJD3
>
>
> Best regards.
>
> -----------------------------------------------------
> Etienne URBAH LAL, Univ Paris-Sud, IN2P3/CNRS
> Bat 200 91898 ORSAY France
> Tel: +33 1 64 46 84 87 Skype: etienne.urbah
> Mob: +33 6 22 30 53 27 mailto:urbah at lal.in2p3.fr
> -----------------------------------------------------
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 5073 bytes
Desc: S/MIME Cryptographic Signature
Url : http://www.ogf.org/pipermail/pgi-wg/attachments/20101209/cb9ec502/attachment.bin
More information about the Pgi-wg
mailing list