[ogsa-bes-wg] Management Functions in BES

[Karl Czajkowski]

On Jul 25, Steven Newhouse modulated:
...
> I don't disagree with your statement.
> 
> 1. Do these exist at the moment for use within the BES specification?
> 

I don't know.  Does the unix "nologin" pattern count (which
administrators requested as an interface to stop job submission on
GRAM)?  My point is not that there would be _one_ standard but that
there are different localized administrative mechanisms which already
exist and are even preferable to yet another service-specific control.

I used the word "profile" because I personally do not expect to find
convergence on one administrative model among the BES constituency...


> 2. Would I want to adopt a whole new set of specifications (assuming the 
> answer to 1 is yes) to turn this off in a 'generic' way?
> 

You tell me. :-) I would want to adopt one, or rather, I would want my
BES implementation to compose w/ the other standards I already must
worry about instead of introducing yet another mechanism.


> >I think it is wrong to burden BES implementations with these kinds of
> >management functions that are specific to a hosting environment. 
> 
> Altering the behaviour of one operation within the service has nothing 
> to do with the hosting environment in my view.
> 

I see this as fundamentally an authorization mechanism, just like the
nologin thing.  Quite likely, the site may want to exempt certain
users and/or classes of jobs from the "do not accept jobs" mode.  In
our Globus Toolkit web service container, authorization is
conveniently handled on a per-operation basis and is often handled by
more generic mechanisms at the message-handling layer rather than by
service-specific code.  (All sorts of variants are possible.)

Also, historically, I think management models like CIM and now WSDM
have tried very hard to identify general lifecycle patterns such as
this "stop accepting work but keep processing" and they are trying to
standardize the administrative models for such things. They may want
to automate the control of such things using workload managers or
other adaptive throttling behaviors. I think a different question is,
can I continue to make do without this extra BES function until such
standards are ready?


karl

-- 
Karl Czajkowski
karlcz at univa.com