Title: Challenges and Opportunities in Grid Middleware
1Challenges and Opportunities in Grid Middleware
Oxford e-Science TAG Meeting January 21 2003
- PTLIU Laboratory for Community Grids
- Geoffrey Fox
- Computer Science, Informatics, Physics
- Indiana University, Bloomington IN 47404
- http//grids.ucs.indiana.edu/ptliupages
gcf_at_indiana.edu
2Fox Comments on UK e-Science
- Excellent Program with coordinated and
application oriented approach - Industry involvement, OGSA, Data Grid, GGF
strategy good - Regional/National centers good idea
- Quality appears generally high and competitive
with international standards - Some risk from hype associated with both Grid
and e-Science leading to unclear technical and
political situation - Both words do not have general agreement
- Suggest defining some safe metrics of success
- Education Grids are an untapped opportunity
3Why discuss Grid Software?
- There is intellectual opportunity in several
areas where there are application requirements
and interesting computer science research issues - Workflow, federation
- There is a need to deploy ROBUST Grids across the
UK (World) to enable cross-country or global
e-Science - Thats where we can ask about level of effort
needed in current activities and if UK (Europe,
USA) should enhance existing projects or start
new synergistic activities - I happen to think there are omissions in current
projects which are nontrivial to retrofit -- for
example - Need for good messaging infrastructure
- Rich Security model supporting different practice
- Peer-to-peer capabilities
4Further Suggestions
- Grid architecture could be addressed in a broader
fashion than just the OSGA (core) level - Add coordination in areas like portals (e.g. use
portlets and develop application web services,
and participate in GCE Research Group) - Although there are separate programming and
Grid Computing Environment groups in GGF, the
focus of both can be considered as programming
the Grid and hence obviously very important - Workflow is just the programming of Web
Services and likely there will be a few distinct
approaches not just XML-based BPEL4WS from
industry - Benchmarks and an international testbed place
to do testing of functionality and performance
seem to be essential
5Grids and e-Science
- There could be more discussion of what is Grid
Software, what is Grid Software Development
Model What do we really need as well as what
are Grid interface standards - What will Globus, IBM, Avaki, Sun, Platform,
Fujitsu, HP, Oracle .. really deliver, will they
interoperate and when will it happen? - Is it true and does it matter that Complete
Globus 3 is an order of magnitude more complex
than Globus 2? - IBM aimed at Enterprise and too complicated for
most real people? - Should one have N Complete Grid systems or an
architecture allowing one to compose your Grid
with components from N different providers? - One must expect different parts of e-Science
community to select different Grid providers
Legion was always object-based (nearer Web
Services than Globus) but Globus wasmore
effective in practice. Maybe robust Avaki Grid
is just what you want? Could you switch from
Globus to Avaki or more realistically
interoperate betweenAvaki and Globus? What
happens if some organization (e.g. industry)
demands better security How do you provide?
6P2P Networks and Grids
- Grids manage and share asynchronous resources in
a rather centralized fashion - Peer-to-peer networks are just like Grids with
different implementations of services like
registration and look-up - P2P Networks like Grids can be built from Web
Services - As typical network delay (gt100ms) large, one can
buid very sophisticated logic into messaging
environment - Publish-subscribe, adaptive routing, QoS,
Security . - P2P Grids naturally federate and support concept
of Personal/Private Grid proposed by US LHC
experimentalists
7Peer to Peer Grid
Peers
Service FacingWeb Service Interfaces
Peers
User FacingWeb Service Interfaces
Peer to Peer Grid of Resources supported by
internal MessageGrid of message or event
brokers Events are just Time-stamped messages
8Users
Portal such as Jetspeed
Hosting Environment
Hosting Environment
GridComputing or ProgrammingEnvironments
Generic Application Services
Web Services
CoreGrid
Core System Services
Resource Grid Services
e.g. DAI compliantdatabase
Resources
9Federation/Interoperability Problem?
- Have a collection of Web Services running in
Grids defined by different suppliers? - Interoperability particular application Web
Service of supplier X can utilize core service
of supplier Y - Federation core service of supplier X can be
integrated with core service of supplier Y to
provide a integration/amalgam that is also a
realization of core service
10OGSI Interface Implications
- OGSI defines the features that all Web Services
must have to be called Grid Services - Using Web Services is equivalent roughly (in
old-style computing lingo) to agreeing on a
common (object-oriented) language - It is equivalent to choosing C or Java or more
precisely agreeing to use CORBA IDL or Java to
define external interfaces (method calls) without
prejudice as to implementation language - For example, having a Grid Information Port is
VERY roughly analogous to agreeing that all
programs have a standard input and a standard
output familiar from UNIX - Agreeing that Grid Information Ports should
communicate XML records (SDE Service Data
Element) is roughly equivalent to saying standard
output will use Fortan namelist syntax - So this is not yet a complete interoperability or
federation specification
11The Grid Onion
- Like most systems, Grids can be thought of
hierarchically with a set of core services
surrounded by other (Web) services with
increasing particularity - The outer levels are likely to be easiest to
agree on with lasting specifications (XML
metadata and Web Service specifications) - Portals are in my opinionan area where
investmentwill have lasting impact asat edge of
architecture - Wise UK investment in OGSA-DAI similarly lasting
- A lot of meta-data specificationactivities can
proceed irrespectiveof inner uncertainity
12Possible Project InterGrids
- New open-source open-process effort developing
Grid software including both the core and
outer-levels of the onion - Key underlying principles
- Support small Grids (university departments,
Private Grids) with easy deployment (e.g. allow
EJB, Oracle etc. but do not require) - Support composability (P2P) and federation with
itself and other Grids - Mitigate risks in e-Science
- Allow innovative computer science research
- Use existing (Avaki, Globus SRB ICENI ..) bits
and pieces if possible encourage such projects
to produce modules useable in other Grid systems
like InterGrids - Good test/benchmark/tutorial material and good
support
13Time Scale and Plan for InterGrids I
- InterGrids has three distinct components that are
synergistic and have some natural overlap - FederatedGrids enabling federation of multiple
OGSA or FGSA compatible Grids - GridLite enabling light weight and personal Grids
that can federate - GridTestbed International compliance,
Performance and 24by7 robustness test environment - Now Set up requirements/planning team Solicit
interest in project gather comments broadly - Report March 2003 allowing call for participants
- Around April/June 2003 Initialize project and
define initial activities, InterGrids
architecture and software-stack Identify
external software likely to be used
14Time Scale and Plan for InterGrids II
- May 2003 Define initial FGSA Federatable Grid
Service Architecture characteristics of Grids
that can be federated in InterGrids - (Obviously differences between OGSA and FGSA
should be minimal and other GGF standards like
GridFTP and OGSA-DAIS must be supported in FGSA) - Around August 2003 Start software development
and development of test applications/environment - Around October 2003 Set-up testing environment
for InterGrids and other Grid Systems - Around mid-2004 Initial (tested) deliverables
- Continually go to GGF, test and gather
requirements - At or Before mid-2006, deliver full tested
software with documentation and defined
interfaces - Include full portal and application Service
generation kit.