09/01/2006 12:28 PM
post5654
|
Re: NorduGrid comments on the GGF UR draft
{context elided...}
> General observations
> ====================
>
> - The document represents a long-awaited attempt at standardization of
> resource usage records. We look forward to implement the format that
> will be agreed by the GGF participants. Still, much work has to be
> done in order to deliver an unambiguous specification that suits
> everybody and accommodates most important aspects.
Thank you for your input, to the document and the process.
> - The document does not give a clear definition of the "usage record",
> which leads to rather confusing statements such as "Record identity
> uniquely defines a record in the usage record" (page 6, definition of
> RecordIdentity).
Additional context has been added at the beginning of the document.
> Furthermore, when talking about the logging & accounting information
> associated to grid jobs it is crucial to give some kind of definition
> of a grid job. Many of the UR properties described in the document are
> ambiguous because it is not clear what constitutes a grid job. An
> example is StarTime. Is the brokering or stagein phase of a grid job
> taken into account?
The document described an atomic record of resource consumption. As such, discussions of "grid jobs", "aggregation", etc
are outside the scope of this document. They may be addressed in V2, if the community feels this is a critical need.
> - The document states that "its main purpose is to outline the basic
> building blocks of the accounting record"
>
> NorduGrid agrees with this scope, we consider the UR rather as a
> logging and accounting record of a grid job rather than simply a
> record of resource consumption. For us the UR should contain
> information not just about resource consumption but also about job
> identity, ownership, status, etc. Maybe it would make sense to rename
> the UR as logging or accounting record?
It is difficult to reach consensus on definitions of "job", "log" and "accounting". For this reason, this recommendation
attempts to identify the smallest unit of resource consumption. Any other expansions of this definition are out of
scope at this level.
> - Many of the attributes are inadequately defined or their meaning is
> deliberately left open. This partly defeats the purpose of the schema
> since it was designed specifically to be a format for exchanging usage
> data over grids but if interpretations of the attributes are different
> at different sites/grids, comparing exchanged data becomes
> complicated. For example the Charge, Status, StartTime, EndTime, base
> properties claim that "The meaning of this charge will be site
> dependent", "the semantic meaning of the status is site dependent" or
> "the value of this property may depend on the queuing system", etc.
The format is meant to facilitate information exchange within a single grid instantiation, which will include
heterogeneous resources. The first step in a unified solution is to unify the atomic data. That is what this
recommendation addresses. Since each grid instantiation will be driven by the policies and practices agreed upon by the
partners in that grid, the recommendation is left sufficiently open to allow common meanings for that instantiation to
be agreed upon.
Meta-grids, grid interoperability, and grids-of-grids are beyond the scope of this version of the recommendation.
> - The GGF UR proposal is still too 'site' or batch system specific; in
> many places the document assumes data exchange among computing centers (sites)
> and not among Grids. Seems that the grid layer is not really
> taken into account, the UR at many places resembles as a data exchange
> format between batch systems (or sites) and not Grids. For example,
> the Current Practices Survey (Appendix A) covers only supercomputing
> centers...
View Full Message
|
|
|