The Grid: What Next? Karim Djemame Web Science Research Group School of Computing
2
Aim of the Talk
• Aim:
Discuss a roadmap for Research & Development and Standardisation for Grid technologies to lead to a globally interoperating Grid for academic research and business use.
3
Grid Vision for … 2020 (1)
• A single interoperable system like the Web today, with general indexing and search capability providing universal access to services through rich semantic descriptions which can be composed into workflows and charged for by a universal accounting system as a utility.
Source: EU Challengers Project
4
Grid Vision for … 2020 (2)
• The Grid will be used equally be academia and business as the Web is today.
Source: EU Challengers Project
5
Grid Vision for … 2020 (3)
• The Grid will provide services including computing power for user programmes, computing applications, storage, and reasoning utilities. This is a natural development driven by early providers such as Amazon and Google who can offer their spare capacity to users now.
Source: EU Challengers Project
6
Grid Vision for 2020: What it Means
• virtualised location• virtualised heterogeneous OS• to provide maximised utilisation on local resources with
outsourced peak demand provision• load balancing• access to large data resources• composition of distributed services into workflows• yellow & white page registries for service advertising• semantically rich description of resources including QoS• user feedback and reputation registries as quality measures• automated management of resources through policies, SLA
and contracts• brokers and agents for management functions• utility accounting for usage.
8
No Roadmap for Grids (yet)
• There is no general agreement yet on the roadmap for Gridsat Internationalat European level.
• There is not even agreement between and within research projects
9
Roadmap for Grids
2006 20182012 20152009
SOKU Apps
SOKU middleware
Web & GRID Services
Ontology Services
OGSA
Web Services
G-Lite
Globus
Unicore
Linux
Windows
Linux-WS
Win-WS
Linux-GS
Note: coloured lines indicate predicted translations to lead to the vision
Courtesy of Keith Jeffery, EU Concertation Meeting, September 2007
10
Roadmap (1): Web and Grid Services
• Convergence of Web and Grid middleware to produce a single universal e-infrastructure for Business Research Other: leisure / entertainment, education,
healthcare, environment management etc
• Started tentatively with WS/OGSA in 2002 but still requires driving forward to meet
requirements
11
Web and Grid Services Convergence: Requirements (2)
• High quality specifications of metadata – with formal syntax and semantics to describe services, with
schema (quality control) associate (descriptive, restrictive) components supportive (e.g. ontologies) components;
• Standards supporting those specifications to provide assurance to software developers of a stable market;
• Standards to provide assurance to end-users of a choice of offered services which interoperate;
• Assurance of non-functional aspects throughout the services ‘stack’:
security, privacy, trust, performance
12
Roadmap (3): OS
• Operating systems support
• Linux (and ideally also Windows) will need to move to a SOA such that their component parts can be dynamically loaded on demand and effectively integrated into the application
• See XtreemOS project efforts Building and Promoting a Linux-based Operating
System to Support Virtual Organizations for Next Generation Grids (2006-2010)
13
Roadmap (4): Middleware
• Marginalisation of Globus and UNICORE is predicted
• Increasing use of a new middleware leading to its evoluation/implementation in a rich SOA
E.g. g-Lite through EGI (European Grid Initiative) See Enabling Grids for E-sciencE project http://www.eu-
egee.org/
• Foundationware: lower middleware to hide the OS and make a uniform platform for resource availability/scheduling
• Middleware: upper middleware to provide commonly required services for interoperation;
• The interfaces require standardisation
14
Roadmap (4): Middleware … but
• There are too many competing interests We won’t see a single Grid middleware
• Interests take a variety of forms Needs: e.g. developing solutions that are suited to
particular patterns of use, commercial interests Competing interests mean that there will be a zoo of Grid
middleware.
• Various middleware will need to co-operate to perform work and also be able to co-exist peacefully when sharing resources.
• The role of standards is to provide a framework that allows
co-existence and encourages cooperation applications to be built on top of interoperability.
15
Roadmap (5): Service Oriented Knowledge Utilities
• SOKU as the end-goal for Application-building Middleware (ideally) OS components
• This achievement depends especially on (again) high quality specifications of metadata
with formal syntax and semantics to describe services
• Key to success Specification, acceptance and standardisation (by
academia and industry) of the ‘envelope’ (i.e. the metadata to describe and control the operation) of a service (i.e. a SOKU) whether it wraps • a source (data, software)• a resource (computer, data store, communications)• an existing service
17
More on the route
• Academic grids will not be sustainable- need to use common middleware as industry to be
economically viable. - must adopt OGSA WS* based software if this is to be the
industry norm, or- must persuade industry to support some special solution by
showing a special market for it
• Key to growth in inter-enterprise WS* Grids will be semantic description of services, service registries and service composition- Including QoS details needed to manage inter-enterprise
relationships- Not supported by existing UDDI registries- Need to provide registries which can support rich service
descriptions
18
More on the route
• Standardisation of the controlled vocabularies, and syntax of the languages to describe services, policies, SLA, contracts well beyond the security policies (SAML and XACML), or the basic agreements (WS-Agreement)
• Need to show the Grid community that the benefits of interoperability outweigh the costs of reduced performance- make service description, advertising and composition usable
by technologists and business people
19
A final note
• Standardisation process takes years, if work is currently a research topic- Do the research- Get it to the standards body in 3 years- May become a standard (if you are lucky) in 6 years
• If the vision is going to be achieved … more questions- Need for regulations and legislation for the governance of
the Grid ? - Should Grid reputation rating services and agencies be
regulated ? - What about service quality or service advertising standards
in an international Grid ? - What is the liability of an agent or a broker - how do courts
resolve such liabilities ?