GGF9
Semantic Grid Workshop
Sunday, Oct. 5, 2003
About 65 persons participated in the
Semantic Grid Workshop of the Semantic Grid Research Group. This
workshop, which is an activity of the GGF Semantic Grid Research
Group, was an opportunity for Semantic Grid practitioners to present their
latest work in the Semantic Grid area, i.e. Grid projects using Semantic Web
technologies. The workshop was also aimed at would-be practitioners who wish to
see the state of play and current practice in this field. The workshop was
open to anyone registered for GGF.
The workshop had four distinct
activities:
- A set of presentations on
Semantic Grid Research topics
- A question and answer session,
which included a discussion of the possible timeline for Semantic Grid
developments
- A collection of concerns and
research topics that the Semantic Grid will need to deal with, as
solicited from the participants as brief notes
- A presentation on a Semantic
Grid Primer - the current major document expected from the Research
Group
Presentations on Current
Semantic Grid Research
- click for the
abstracts
Presentations were condensed into written papers, which will be combined into
an informational document for GGF.
A Question and Answer Session
The Q&A session during the workshop was a lively encounter between many
members of the audience.
The opening thread of conversation concerned how to create well-engineered
and affordable ontologies. There was clearly concern that there has been
insufficient experience in the Semantic Grid community to understand how to
create ontologies that scale well or that are maintainable. Indeed, some
participants felt that exclusive concentration on ontologies might not be the
optimum road for rapid expansion of this technology. One suggestion was
that entity-element triples might be an easier way to start, since these are
familiar to database practitioners. A second suggestion was that a
description of ontology creation and use might be very helpful in creating a
greater familiarity with the technology.
A second thread of discussion was that semantic approaches might be useful in
dealing with services - both web services and grid services. If that is
the case, one of the research areas was likely to be creation and use of tools
(which we might call 'Integration Engines') to speed semantic grid work.
On the other hand, some participants felt that it was up to services to provide
a description of any semantic knowledge they might use or provide.
A third thread of discussion was tied to a discussion of how to develop good
ontologies.
These threads were extended in the
collection of challenges and research topics, described in the next section of
these minutes.
A Collection of Concerns and
Research Topics
As part of this
workshop, the organizers requested each participant to write down suggested
concerns and research topics on small, post-it notes that were then collected
from the participants the end of the Q&A session. The concerns
were as follows:
1. How will semantic
annotation performance scale? When everything is distributed, can we perform all updates in
a finite time?
2.
Extensibility, etc.
-
Extensibility
and interoperability of different semantic grids -> unification of ontologies
-
Building optimal
workflows “at run time”
-
Security aspects
3.
Mappings
-
How can we
facilitate the mapping of (several, many) metadata schemas independently
developed by scientific users who realize after the fact that they should share
their data? Or publish it in a
repository using another metadata schema?
-
Can or should
this be done without referring to upper ontology?
4.
We didn’t agree on an ontology before entering the room –
we map and remap between personal ontologies as we go. We need to build semantic grid tools that allow
dynamic/evolving mapping to avoid the costs of complete, up front agreement (and
avoiding dealing with complete models when approximations suffice).
5.
Translation
-
Extracting the
semantics or groundings of all the scripts used to glue various species of the
grid together
-
With some
vocabularies and ontologies, is there some notion of an ontology translation or
matching?
6.
Challenges
-
Lifecycle
ontology management (Using Date matching can help)
-
Semantic grid
should offer tools for application design, mapping, execution enhanced with
respect to level => Build Semantic Grid (programming tools)
-
Tools for the
Semantic Grid:
i.
Application design -> component based one (?)
ii.
Distributed data mining could enhance knowledge acquisition and ontology
building, such as workflow behavior
7.
How do we keep
the semantic annotations – metadata, services – up to date (versioning,
changes, conformance levels, etc.)?
8.
Concerns
-
Cost and
difficulty of acquiring semantic information
-
Tools and tools
for ‘domain’ experts are two or three … actually 4 challenges (happy to
carry a passenger who doesn’t have enough problems of their own)
9.
Frustrations
-
Multiple
ontologies – get over it!
-
Ontology Jag …
Need a want…
10.
Distributed
Ontologies
-
How to
semantically define a network tuning service.
-
Since provided
data is locale specific, is an ontology needed?
Multiple services will not be available because only one service will
provide information per locale. If
more than 1 service exists, similarity is moot.
Service with the most recent data will be the one to choose.
11.
Challenges
-
Providing a well
engineered process that creates and maintains ontologies that cover a sufficient
amount of information to cover the interlocking domains of grid application.
This cannot depend on humans alone.
-
Being able to
understand the above ontologies at that scale to be confident that they lead to
appropriate process inferences.
12.
Challenges
-
Proliferation of
metadata, especially incorrect or invalid data
-
Security
-
Standardization
13.
How will the GRID
community encourage/support collaborative ontology design and modification?
14.
Challenges
-
Correlation of
ontologies
-
Consistency
management
15.
Are there any
implementations on the human behavior on-going or in plan?
16.
Challenges
-
Data integration
from multiple heterogeneous data sources, such as resource description
information
-
Scalability and
performance of ontology-based/knowledge-based reasoner
17.
Challenges
-
Grid Resource
Ontology
-
Application
tools/API’s,/Services
18.
Several speakers
mentioned the desirability of distributed ontologies.
To enable this, what needs to be done about
-
Tools and mapping
ontologies
-
Community process
19.
What tools can be
offered for other grid portals that do not use the same infrastructures that
facilitate incorporating the semantic grid into their portal with local
resources included (How does the semantic grid become a node in a large,
upstream grid).
20.
How do we create
all the required semantic annotations in an environment rich in data object
types, data formats, and complex simulation algorithms?
21.
Challenges
-
Achieving at
least an initial characterization of the kinds of metadata, for example
domain-level versus provenance, and understanding which impinge on file/data
content
-
Building semantic
characterizations of workflows that support joint manual and automated
construction, to support reuse, discovery and maintenance/repair, that has
well-defined links with the domain ‘data’ ontology
22.
Validation that
workflow and onotology semantics satisfy a wide range of science and engineering
work practices and resource discovery needs
23.
Challenges
-
Ontology for the
grid itself
-
Performance at
run time
-
Complexity
The Semantic Grid Primer
The Semantic Grid Primer is
currently expected to be the major document created by the Research Group.
Carole Goble provided a useful presentation on this document and the approach
being taken to put it together:
As a result of this discussion, the
Secretary of the Semantic Grid Research Group volunteered to produce a marketing
plan for the Primer.