[ogsa-wg] Thoughts on extensions mechanisms for the HPC profile work

[Marvin Theimer]

Hi;

 

I certainly agree that you will want to be able to define more complex
expressions than just lists of identifiers and that you will want to be
able to create various kinds of aggregations - including workflows -
that persist beyond a single client-service interaction.  The key point
I was after was to observe that there are situations where a client will
want to specify multiple entities in a request and that therefore we
need a way of accommodating such requests.  (Note that you sometimes
might want to name multiple aggregate entities as well - e.g. to cancel
all the workflows that you've currently got running.)  So the main thing
I want to avoid is ending up with a purely "object-oriented" design in
which all stateful entities are accessed strictly via separate messages
sent only to their associated EPRs is not sufficient.

 

Another thing to keep in mind is that not all interactions with
logically aggregated entities require a persistent connection.  Suppose
I want to periodically query the status of a particular set of jobs.  If
I periodically send across a list of job IDs then the scheduler will
assemble the current status information for each listed job and return
that information to me.  If I have to first ask the scheduler to create
a stateful object representing that list of jobs, so that I can then
send query messages to the list object, then the scheduler has to do
substantially more work for no real benefit.  In particular, it still
has to assemble and return the current state information for each job in
the list for each query.  But now it also has to maintain a stateful
list object as well, including all the state and management duties that
are associated with that.  Worse yet, you now get to deal with all the
failure modes that a stateful connection can exhibit in the face of
client and server crashes, as well as network partitions.  So, whereas I
agree that creating stateful representations of more complex entities,
such as workflows, is definitely a case to be supported, I would
disagree with the notion that every aggregate collection - no matter how
ephemeral - should be dealt with by means of reified aggregation
objects.

 

Moving on, you are right that an abstract (opaque) name is not very
useful without knowing which service(s) can understand its meaning.  But
that's not really what I'm proposing: When you get back an abstract name
from a service the understanding is that you can use the name when
interacting with that service (or other services that you explicitly
know will understand it).  You are expected to remember that binding.

 

Personally I like WS-names for this - and other - reasons.  Because the
abstract name - the true name, if you will, rather than a potentially
ephemeral address for where to send messages to - is explicit you can do
all kinds of useful things, including extracting an efficient
representation of a bunch of them.

 

I'm not sure what you mean with your suggestion of creating a single
WS-Name that can be used to query a bunch of different EPRs' entities.
The benefits I'm talking about come from not involving the server in any
explicit aggregation "creation" operations.  If you mean to make the
notion of extracting abstract names from a bunch of WS-names and then
using them in an array operation that is sent to the same EPR, then I'm
all for it.

 

Marvin.

 

________________________________

From: owner-ogsa-wg at ggf.org [mailto:owner-ogsa-wg at ggf.org] On Behalf Of
Dave Berry
Sent: Saturday, April 29, 2006 8:13 AM
To: Marvin Theimer; ogsa-wg at ggf.org
Subject: RE: [ogsa-wg] Thoughts on extensions mechanisms for the HPC
profile work

 

Hi Marvin,

 

It seems strange to me to limit yourself to a list of identifiers as a
means of interacting with large numbers of jobs.  Wouldn't it be more
flexible to allow more complex expressions, such as "all jobs submitted
by Marvin that use more than 4 processors and are currently running" (or
whatever).  You could build an appropriate information model, map it to
a data model and provide a corresponding query language that clients
could use to request the information they need.


I agree that this is a service-oriented approach rather than a
resource-oriented approach.  However, having set up a particular query,
you might want to repeatedly interact with the current state of that
query.  So it might make sense to register the query with the service
and return an EPR that can be used to get the current set of results.

 

I don't see much point in passing back an abstract name (e.g. a UUID) on
its own.  Without a reference to a resolution mechanism, clients won't
be able to make much use of it.  This is why URL's and URI's have been
so successful; they include enough information for any client to know
which mechanism to use to resolve them.  This seems to be one advantage
of the WS-Name proposal; it includes an abstract name with a reference
to a resolution mechanism. 

 

Is it still the case that BES containers are allowed (or even
encouraged) to return WS-Names?

 

Would it be useful to have a means for composing WS-Name EPRs that use
the same resolution mechanism in order to make a single WS-Name that can
be used to query all of them?

 

Dave.

 

 

 

 

-----Original Message-----
From: owner-ogsa-wg at ggf.org [mailto:owner-ogsa-wg at ggf.org] On Behalf Of
Marvin Theimer
Sent: 29 April 2006 03:06
To: ogsa-wg at ggf.org
Subject: [ogsa-wg] Thoughts on extensions mechanisms for the HPC profile
work

	1.      *        Support for array operations and other forms of
batching.

	2.      *        When 1000's of jobs are involved the efficiency
gains of employing array operations for things like queries or abort
requests are too significant to ignore.  Hence a model in which every
job must be interacted with on a strictly individual basis via an EPR is
arguably unacceptable.

	3.      *        One approach would be to simply add array
operations alongside the corresponding individual operations, so that
one can selectively interact with jobs (as well as things like data
files) in either an "object-oriented" fashion or in "bulk-array"
fashion.

	One could observe that the array operations enable the
corresponding individual operations as a trivial special case, but this
would arguably violate the principle of defining a minimalist base case
and then employing only extensions (rather than replacements).

	4.      *        Array operations are an example or a
service-oriented rather than a resource-oriented form of interaction:
clients send a single request to a job scheduler (service) that refers
to an array of many resources, such as jobs.  This raises the question
of whether things like jobs should be referred to via EPRs or via unique
"abstract" names that are independent of any given service's contact
address.  At a high level, the choice is unimportant since the client
submitting an array operation request is simply using either one as a
unique (and opaque) identifier for the relevant resource.  On a
pragmatic level one might argue that abstract names are easier and more
efficient to deal with than EPRs since the receiving scheduler will need
to parse EPRs to extract what is essentially the abstract name for each
resource.  (Using arrays of abstract names rather than arrays of EPRs is
also more efficient from a size point-of-view.)

	5.      *        If abstract names are used in array operations
then it will necessary that individual operations return the abstract
name and not just an EPR for a given resource, such as a job.  If this
approach is chosen then this implies that the base case design and
implementation must return abstract names and not just EPRs for things
like jobs.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.ogf.org/pipermail/ogsa-wg/attachments/20060501/785d9a8c/attachment.htm