[ur-wg] Meeting Notes
Rosario Michael Piro
piro at to.infn.it
Thu May 11 07:41:21 CDT 2006
Matthew Ford wrote:
> Quoting "Donal K. Fellows" <donal.k.fellows at manchester.ac.uk>:
>
>> Future version to extend to more complex things than batch system job
>> accounting.
>
>
> But I think think we have this now! Maybe I'm reading too much into the
> term
> "extend". It's much more a change of method and clarification that I would
> like to see. As you move through-out the document it becomes increasingly
> generalised. By section ten, I believe the current specs clear
> intention is to
> be able to account much more than a batch job. However it's not with out
> problem.
>
It is true that the general structure of the UR and most of it's
elements allow for a more general case, but I believe that that would
not be enough. Storage accounting, for example, would most probably
require additional elements such as FileIdentity/FileName. Other
elements (Disk) are already in place. I think the usage-related elements
(network, disk, memory, etc.) already provide most of what would be
needed, but the identity-related elements are still too batch
job-specific. We should concentrate concentrate generalization efforts
first on these.
> Some thoughts and quotes:
>
> - Job Level Accounting from the Original Spec is supposed to work at the
> batch
> job level. It is restricted form of the usage of the base properties of
> section 3 or later the common properties of section 10.
>
> From Sec.4: "Job level accounting reports accounting data at the job
> level. PBS
> and LoadLeveller. This type of usage record MAY contain any of the base
> properties. This type of usage record MUST contain at least one of the
> following properties:o LocalJobId o GlobalJobId"
>
> - Aggregate Accounting is specifically allowed, and integral to the
> current doc.
I still believe that what is missing is an explicit differentiation
between aggregate records and non-aggregate records (this might also be
a simple attribute type="aggregate" to the UsageRecord element!?). An
aggregate record should at least clearly state that it is aggregate and
this statement should be unambiguosly defined, such that each
UR-"consumer" (whether a user, or a RUS) can undoubtfully distinguish
between these records). An aggregate record might (or should) also allow
multiple JobIdentity elements to be present.
>
> From Sec.5: "Aggregate accounting reports the accounting data in aggregate
> (summarized form). Aggregate accounting MAY contain any of base properties
> listed in this document,"
>
> From Introduction: "Aggregation: Most sites have indicated that job-level
> aggregation is sufficient for their current needs. Where sites aggregate or
> decompose to other levels of usage tracking, provisions are
> available within the schema to accommodate."
>
> From Sec.3: "Base Properties: The following is a list of base
> properties that
> define common usage record requirements for both job level and aggregate
> properties."
>
> - The Global definitions in sections 7,8 and 9 clearly intend to setup a
> generic
> framework in which most types of usage can can be accounted.
> + Sec.7 sets-up Global Elements to allow substitution for inclusion
> into other
> schemas.
> + Sec.8. sets-up Global attributes sets up generic attributes which
> maintain
> meaning independent of context. i.e, forcing common usage of units.
> + Sec.9. specifically sets-up a *generic* usage record
>
> From Sec.7 "The global definitions are those that may be used
> compliantly with
> this specification as root elements for an XML document or which may be
> used as
> extension points within or included within other XML Schema definitions."
>
> From Sec.8 "Global attributes, and attribute groups, are attributes
> that are
> common to many of the element definitions contained in this
> specification. A
> global definition of these maintains standard semantic meaning within each
> context that they are used."
>
> From Sec.9 "UsageRecordType Complex Type: This complex type definition
> establishes the structure of the *generic* usage record."
>
> - Section 10, Usage Record Common Properties is an obivious attempt to
> generalise the base proprieties of section 3. The problem is that not
> all the
> base properties of section three are appropriate as Common Properties for a
> *generic* record. Further as the bulk of section 3 is seemingly based
> on batch
> level reporting, the scope of the section is serverely limited.
>
> From Sec.10 "10. Usage Record Common Properties: the complete
> specifications for
> each element that MAY appear in the *generic* usage record are presented
> below."
>
>>
>> Stick with using email for resolving issues; not worth changing
>> technologies half way through.
>>
>> Short-term Timeline:
>> Mailing list discussions to close: 30 June
>> Consensus: 15 July
>> Revised document: 15 August
>> Submit to editor: 1 September
>>
>> Three-way joint meeting: OGSA RUS UR
>>
>> Call for use cases and reference implementations
>>
>> Telecons
>> ========
>> Keep going. Biweekly. One hour. 1300 UTC. First on 25 May.
>> ----------------------------------
>>
>> Donal Fellows.
>>
>>
>
--
-------------------------------------
Rosario Piro (piro at to.infn.it)
http://www.to.infn.it/~piro/
-------------------------------------
Istituto Nazionale di Fisica Nucleare
Sezione di Torino
-------------------------------------
National Insitute for Nuclear Physics
Section of Turin, Italy
-------------------------------------
More information about the ur-wg
mailing list