Re: Comments of PROV-DM document (Section 2.1 and 3)

Hi Luc,
>The notion of event introduced in the document is different. It should be
understood as "a transition that
>changes the state of the system".

The current definition of event from PROV-DM document: "In our
conceptualization of the world, instantaneous events, or *events* for short,
happen in the world, which mark changes in the world, in its activities, and
in its things."

Given that instantaneous is *always* defined with respect to a unit of time:
Instantaneous: "done, occurring, or acting without any perceptible duration
of time" - Webster dictionary

and there is still no absolute unit of time corresponding to "instant" [1],
each application (provenance and others) use a relative notion of
"instantaneous".

Hence, what is an "instantaneous event" for provenance application A maybe a
"long-running event" (aka PE in your language) for another provenance
application according to what each application considers to be
"instantaneous" - 1 second, 1 millisecond, 1 nanosecond etc.

A scientist can assert that a workflow instance "mappingGenetoProtein"
"started" at 11:00am US ET and "ended" at 11:15amET - so
"mappingGenetoProtein_start" and "mappingGenetoProtein_end" are events
according to your current definition.

But, a programmer views the "mappingGenetoProtein_start" as a PE since it
has the following events (according to your definition) - "load gene
sequence" at 11:00:05am ET, "create packet with gene sequence and header
information" at 11:00:20am ET, and "send request to NCBI BLAST website at
11:00:50am ET etc.

It is important to note that "nothing" happened between 11:00:00am ET and
11:00:04am ET, but for the researcher the event of
"mappingGenetoProtein_start" occurs at 11:00am ET since the granularity of
time measurement for the researcher is hours and minutes, not seconds.

All the above listed events satisfy your current definition of event:
"instantaneous" "mark changes in the world, in its activities, and in its
things."

Hence, to paraphrase "one person's data is another person's metadata", the
*current definition* of event allows it to be interpreted as "one person's
event is another person's PE".

Overall, there does not seem to be any need for this WG to mandate that an
event has to be instantaneous (and not be able to define what is
instantaneous). We are using events to enable ordering - allowing events to
span time duration does not negate this capability. Each application can
decide on the granularity of time to be associated with an event and still
use events for relative ordering.

What do you think?

> You come to this discussion with a different meaning of event, and try to
shoe horn it into the notion
>of event in the model.
Unfortunately Luc, just the "event" page on Wikipedia lists 19 meanings of
the term "event" [2] - there are many more definitions of "events" in
computer science - from system events to user events (basis of event-driven
programming). The issue is not of shoehorning or unshoding - but a better
understanding of the issues.

> We do not have a containment relationship for PEs.  Whether we want one is
open to debate.
I think it is necessary - a workflow instance PE may contain many "atomics"
PEs.

Thanks.

Best,
Satya


[1] http://en.wikipedia.org/wiki/Instant
[2] http://en.wikipedia.org/wiki/Event

On Wed, Oct 12, 2011 at 3:48 AM, Luc Moreau <L.Moreau@ecs.soton.ac.uk>wrote:

>
> Hi Satya,
>
> What you are describing here is various activities (which we represent as
> process execution expressions).
> The notion of event introduced in the document is different. It should be
> understood as "a transition that
> changes the state of the system".  Start of a PE and end of a PE are
> examples of such transitions/events.
>
> You come to this discussion with a different meaning of event, and try to
> shoe horn it into the notion
> of event in the model. It does not work.  The examples of what you describe
> as events should be mapped
> to PEs.
>
> We do not have a containment relationship for PEs.  Whether we want one is
> open to debate.
>
> Luc
>
>
> On 10/11/2011 11:39 PM, Satya Sahoo wrote:
>
>> We briefly discussed the issue of ordering of events during the PROV-O/M
>> call yesterday and the example we discussed may be relevant - an event
>> "issuance of a traffic ticket to X" can be viewed as instantaneous (total of
>> traffic tickets issued per year in City A) or stated to have a duration
>> (10mins). Further, the event may contain events as "ran the stop sign",
>> "handed over license", "signed the ticket" etc.
>>
>> We should be able to model all the above set of events in PROV.
>>
>
> --
> Professor Luc Moreau
> Electronics and Computer Science   tel:   +44 23 8059 4487
> University of Southampton          fax:   +44 23 8059 2865
> Southampton SO17 1BJ               email: l.moreau@ecs.soton.ac.uk
> United Kingdom                     http://www.ecs.soton.ac.uk/~**lavm<http://www.ecs.soton.ac.uk/~lavm>
>
>