Feedback from the GLUE schema person
I've been working with...to factor into our info model discussions.
Appears additional collaboration with
Sergio is welcome.
Ellen
Ellen Stokes
STSM, Grid Computing
TC Member, IBM Academy of Technology
Austin, Texas
tl 678 0552 outside +1 512 838 0552
stokese@us.ibm.com
----- Forwarded by Ellen
Stokes/Austin/IBM on 07/10/2006 04:11 PM -----
Sergio Andreozzi <sergio.andreozzi@cnaf.infn.it>
06/20/2006 08:16 AM
To
Ellen Stokes/Austin/IBM@IBMUS
cc
Laurence Field <Laurence.Field@cern.ch>
Subject
Re: info model XML representation
Dear Ellen,
I read the document and found it interesting. As regards the main
question (which of the two proposal is preferred), I have not a clear
answer yet since this depends on the final goal of the paper. Below you
can find my considerations:
Some preliminary statement:
1. the GLUE Schema models Grid resource characteristics
2. the JDL is used to describe a job request; requirements on resources
can be expressed based on GLUE Schema-related attributes
3. JSDL models a job description, hence it contains requirements on
resource characteristics; the requirements are expressed over a set of
characteristics inspired by CIM; this set of attributes can be
"isolated" in a Resource Requirement Language (RRL)
so the proportion: JSD : JSDL = GLUE Schema : RRL
The document describes a solution to both model resource characteristics
and requirements on resources. In order to better position the paper, it
would be useful to state:
1. relationship between the proposal for describing resource
requirements and the JSDL Resources Elements; Figure 1 can help in
understanding, nevertheless it would be better to write it more explicitly
2. better state the goal of the modeling, in the sense that it should be
more clear if the purpose of the document is to describe a modeling for
enabling a resource selection process; (also in the abstract);
a possible discussion on the matchmaking approaches using this modeling
could also fit the paper;
3. it is generally more acceptable to write about "a Grid" or
"the
Grids" instead of "the grid", since there is no a unique
Grid as it was
envisioned in the early years of this paradigm
4. about the two possible approaches, at a first glance, I do not lean
to a particular approach; some comments below:
Proposal I - Using XML typing and extensibility
pro:
* enables better data quality enforcement since it relies on XML Schema
data model
* characteristics modeled as XML elements (instead of attributes) give
more flexibility in defining complex type
neutral: there is no a symmetric approach in the description of
resources and activities
consideration about the node entity; considering the GLUE Schema experience:
- resources are managed by batch systems and are organized in pools
exposed to the Grid layer
- for scalability reasons, there is no description of single nodes, but
description of homogeneous set of nodes, i.e., homogeneous execution
environments for applications; this implies that only static
characteristics are modeled and advertised (e.g., TotalMainMemory is ok,
FreeMainMemory is ko); dynamic attributes can be added if they refer to
aggregation (e.g., MaxFreeMainMemory, that is the highest free main
memory among the nodes present)
- an important aspect is the service differentiation, that is, do all
Grid users have the same possibility to access a resource or not? this
is currently a hot topic and the solutions being investigated adopt the
creation of different shares/priorities assignable to different
groups/roles hold by users; this means that the resources are the same,
but the selection process should be able to consider the QoS aspect;
these aspects are important in a resource selection process
Proposal II - Using typing outside of XML
this approach is schema-free and symmetric (a la ClassAd);
One more general comment. An important requirement for a selection
system is to be able to express preferences that are used to choose the
best resource among the available ones that matched the
hard-constraints. This topic was the core of my PhD thesis that you
might be interested looking at:
(http://www.cs.unibo.it/pub/TR/UBLCS/ABSTRACTS/2006.bib?ncstrl.cabernet//BOLOGNA-UBLCS-2006-02)
For resource description, the undertaken approach is close to your
proposal I; for requirements and preferences, I've defined a tailored
language inspired by XQuery (it is called XMatch).
I would like to have your feedback about these considerations. Consider
also the possibility of writing a joint paper.
Cheers, Sergio
Laurence Field wrote:
> Hi Ellen,
>
> I think that Sergio is the correct person to comment on this. I
have
> cc'd him,
>
> Laurence
>
> Ellen Stokes wrote:
>
>>
>> Laurence,
>>
>> As part of the information model direction that the resource
>> management design team is pulling together (part of the GGF OGSA
>> workgroup), we are considering 2 alternatives for the XML
>> representation of an information model. We've discussed
the pros and
>> cons of each proposal, but have come to no conclusions or
>> recommendations. So, we've decided to solicit the opinions
of groups
>> that use information models and have an XML representation. And
the
>> EGEE's G-lite work is one such group.
>> I'm not sure if you're the right person to work with to solicit
an
>> opinion on XML representation. If not, please let me know
who I
>> should talk with.
>>
>> I've attached the document. Section 3 proposes the overall
direction
>> for modeling resources in the grid. Section 4 provides an
example
>> (take the example at face value - it's just an example so don't
worry
>> about details of the reousrces modeled) and proposes 2 alternative
XML
>> representations. I'd like to understand G-lite's opinion
on which
>> representation should be used to model resources in the grid and
why.
>>
>>
>>
>> Thanks.
>> Ellen
>>
>> Ellen Stokes
>> STSM, Grid Computing
>> TC Member, IBM Academy of Technology
>> Austin, Texas
>> tl 678 0552 outside +1 512 838 0552
>> stokese@us.ibm.com
>
>