Transcript Document

Grid Technology B
Different Flavors of
Grids
CERN Geneva
April 1-3 2003
Geoffrey Fox
Community Grids Lab
Indiana University
[email protected]
Different Types of Grids
•
•
•
•
Compute and File-oriented Grids
“Internet Computing” Grids (Desktop Grids)
Peer-to-peer Grids
Information Grids: to distinguish between
File, database and “Perl Filter” based Grids
• Semantic Grids
• Integrated (Hybrid, Complexity) Grids
– Bio and Geocomplexity
• Campus Grids
• Enterprise Grids
Compute and File-oriented Grids
• Different Grids have different structures
• Compute/File oriented Grids are well represented by
“production part of particle physics” either in
– Monte Carlo
– Production of Data Summary Tapes
• This is nearer the “Globus GT2” rather than the “Web
Service” vision of the Grid
• Strongly supported of course by EDG (European Data
Grid and Trillium project in the US (Virtual Data
Toolkit)
• Physics Analysis phase of particle physics requires
more collaboration and is more dynamic

What do HEP experiments want to do
on the GRID in the long term ?
Production:
n
n
n
n
n
Simulation (Monte Carlo generators).
Reconstruction (including detector geometry …).
Event Mixing (bit wise superposition of Signal and Backgrounds).
Reprocessing (Refinement, improved reconstruction data production).
Production (production of AODs and ESDs starting from Raw data).
n

Very organized activity, generally centrally managed by prod teams
Physics analysis:
n
n
Searches for specific event signatures or particle types.
(data access can be very sparse, perhaps on the order of one event out of each
million).
Measurement of inclusive and exclusive cross sections for a given physics
channel – Measurement of relevant kinematical quantities
n
n
I/O not feasible to organize the input data in a convenient fashion unless
one constructs new files containing the selected events .
the activities are also uncoordinated (not planned in advance) and (often)
iterative.
EDG “Compute/File” Grid Work
Packages
• WP1: Work Load (Resource) Management System
• WP2: Data (Replication/Caching) Management
• WP3: Grid Monitoring / Grid Information Systems
(general meta-data lookup
• WP4: Fabric Management (software etc. on cluster)
• WP5: Storage Element (Grid Interface to mass
storage)
• WP6: Security
• WP7: Network Monitoring
Compute/File Grid Requirements I
• Called Data Grid by Globus team
• Terabytes or petabytes of data
– Often read-only data, “published” by experiments
• Large data storage and computational resources
shared by researchers around the world
– Distinct administrative domains
– Respect local and global policies governing how resources
may be used
• Access raw experimental data
• Run simulations and analysis to create “derived” data
products
Compute/File Grid Requirements II
• Locate data
– Record and query for existence of data
• Data access based on metadata
– High-level attributes of data
• Support high-speed, reliable data movement
– E.g., for efficient movement of large experimental data
sets
• Support flexible data access
– e.g., databases , hierarchical data formats (HDF),
aggregation of small objects
• Data Filtering
– Process data at storage system before transferring
Compute/File Grid Requirements III
• Planning, scheduling and monitoring execution of
data requests and computations
• Management of data replication
– Register and query for replicas
– Select the best replica for a data transfer
• Security
– Protect data on storage systems
– Support secure data transfers
– Protect knowledge about existence of data
• Virtual data
– Desired data may be stored on a storage system
(“materialized”) or created on demand
Functional View of Compute/File Grid
Application
Metadata Service
Planner:
Data location,
Replica selection,
Selection of compute
and storage nodes
Replica Location
Service
Information Services
Location based on
data attributes
Location of one or
more physical replicas
State of grid resources,
performance measurements
and predictions
Security and Policy
Executor:
Initiates
data transfers and
computations
Data Movement
Data Access
Compute Resources
Storage Resources
COLLECTIVE 2:
SERVICES
SPECIFIC TO
APPLICATION
DOMAIN OR
VIRTUAL ORG.
FABRIC
CONNECTIVITY
COLLECTIVE 1:
GENERAL
SERVICES FOR
COORDINATING
RESOURCE:
MULTIPLE
SHARING SINGLE
RESOURCES
RESOURCES
COLLECTIVE
Layered C/F Grid Architecture
Request
Interpretation
and Planning
Services
Data
Transport
Services
Data Access
Protocol or
Service
Communication
Protocols (e.g.,
TCP/IP stack)
Storage
systems
Workflow or
Request
Management
Services
Data
Federation
Services
Storage
Resource
Management
ApplicationSpecific Data
Discovery Services
Data Filtering or
Transformation
Services
Data Filtering or
Transformation
Services
Authentication and
Authorization
Protocols (e.g., GSI)
Compute
Systems
Networks
Community
Authorization
Services
General Data
Discovery
Services
Database
Management
Services
Consistency Services
(e.g., Update Subscription,
Versioning, Master Copies)
Storage
Management
(Brokering)
Compute
Scheduling
(Brokering)
Compute
Resource
Management
Monitoring/
Auditing
Services
Resource
Monitoring/
Auditing
C/F Grid Architecture I
(from the bottom up)
• Fabric Layer
– Storage systems
– Compute systems
– Networks
• Connectivity Layer
– Communication protocols (e.g., TCP/IP protocol
stack)
– Authentication and Authorization protocols (e.g.,
GSI)
C/F Grid Architecture II
• Resource Layer: sharing single resources
– Data Access Protocol or Service (e.g., Globus gridftp)
– Storage Resource Management (e.g., SRM/DRM/HRM
from Lawrence Berkeley Lab)
– Data Filtering or Transformation Services (e.g.,
DataCutter from Ohio State University)
– Database Management Services (e.g., local RDBMS)
– Compute Resource Management Services (e.g., local
supercomputer scheduler)
– Resource Monitoring/Auditing Service
C/F Grid Architecture III
• Collective 1 Layer: General Services for Coordinating
Multiple Resources
– Data Transport Services (e.g., Globus Reliable File Transfer
and Multiple File Transfer Service from LBNL)
– Data Federation Services
– Data filtering or Transformation Service (e.g., Active ProxyG
from Ohio State University)
– General Data Discovery Services (e.g., Globus Replica
Location Service and Globus Metadata Catalog Service)
– Storage management/brokering
– Compute management/brokering (e.g., Condor from
University of Wisconsin, Madison)
– Monitoring/auditing service
C/F Grid Architecture IV
• Collective 2 Layer: Services for Coordinating Multiple
Resources that are Specific to an Application Domain
or a Virtual Organization
– Request Interpretation and Planning Services (e.g., Globus
Chimera and Pegasus for Physics Applications and Condor
DAGMan)
– Workflow management service (e.g., Globus Pegasus)
– Application-Specific Data Discovery Services (e.g., Earth
Systems Grid Metadata Catalog)
– Community Authorization service (e.g., Globus CAS)
– Consistency Services with varying levels of consistency,
including data versioning, subscription, distributed file
systems or distributed databases
Composing These Services To
Provide Higher-Level Functionality
• For example, a Grid File System might compose:
– Fabric layer: storage components, compute elements
– Connectivity layer: security and communication
protocols
– Resource layer: data access protocols or services and
storage resource management
– Collective layers: transport and discovery services,
collective storage management, monitoring and
auditing, authorization and consistency services
Peers
Peer to Peer Network
User
Service
Resource
Routing
User
Service
Resource
Routing
User
Service
Resource
Routing
Peers are Jacks of all Trades linked to “all” peers in community
Typically Integrated Clients Servers and Resources
User
Service
Resource
Routing
User
Service
Resource
Routing
User
Service
Resource
Routing
Peer to Peer (Hybrid) Grid
User
Service
Resource
Routing
User
Service
Resource
Routing
NB Routing
User
Service
Resource
Routing
User
Service
Resource
Routing
User
Service
Resource
Routing
Services
Dynamic
Message or
Event
Routing from
Peers or
Servers
User
Service
Resource
Routing
A democratic organization
Peers
Database
Database
Service Facing
Web Service Interfaces
Event/
Message
Brokers
Event/
Message
Brokers
Event/
Message
Brokers
Peer to Peer Grid
Peers
User Facing
Web Service Interfaces
Chapter 18 and 19 Grid Book
Peer to Peer Grid
Entropia: Desktop Grid
l
l
l
Entropia (chapter 12 of book), United Devices,
Parabon, SETI@Home etc. have demonstrated
“internet Computing” or Desktop Grid very succesfully
Used to be called peer-to-peer computing but that fell
out of favor due to Napster’s bad name
Condor has similar types of utility but Entropia
optimized for
– Huge number of clients
– Providing a secure “sandbox”
for application to run in
which guarantees that
application will not
harm client
Scaling of Entropia Application
Entropia Architecture
Application Execution on the
Entropia System. End-user submits
computation to Job Management
(1). The Job Manager breaks up the
computation into many independent
“subjobs” (2) and submits the
subjobs to the resource scheduler.
In the mean time, the available
resources of a client are periodically
reported to the Node Manager (a)
that informs the Subjob Scheduler
(b) using the resource descriptions.
The Subjob Scheduler matches the
computation needs with the
available resources (3) and
schedules the computation to be
executed by the clients (4,5,6).
Results of the computation are sent
to the Job Manager (7), put
together, and handed back to the
end-user (8).
Information Grids I
• Actually nearly all Grids consist of composing access
to data with processing of that data in some computer
program
• In Compute/File Grids (Data Grids for Globus), one
naturally allowed database access from programs
although in some cases dominant access is to files
• In Information Grids, we consider access to databases
but view of course files as a special case of databases
• Real difference is what tier we are looking at:
– Compute/File Grids are looking at “backend
resources”
– Information Grids are looking at “middle tier”
because typically data volumes are not large
enough to stress typical middle-tier mechanisms
Information Grids II
• Should use Middle tier where possible and adopt
hybrid model with control always in middle tier and
using backend only where needed
– This would require reworking a lot of tools e.g.
Condor should schedule services not jobs
• Most programming models either specify “program
view” or “service view” and do not separate
– Developments like GT3 will allow changes but it will take a
long time before key tools are implemented in hybrid mode
• Note Bioinformatics and many other Information Grids
only require service view
– These applications have in UK e-Science started with “Web
Service” and not “Globus” view
User
Services
System
Services
Grid
Computing
Environments
Portal
Services
System
Services
Application
Service View
Service
Middleware
System
Services
Program View
System
Services
System
Services
Raw (HPC)
Resources
“Core”
Grid
Database
OGSA-DAI
(Malcolm Atkinson Edinburgh)
UK e-Science Grid Core Programme
Development of Data Access and Integration Services for OGSA
http://umbriel.dcs.gla.ac.uk/NeSC/general/projects/OGSA_DAI
- Access to XML Databases - Access to Relational Databases - Distributed Query Processing - XML Schema Support for e-Science -
DAI Key Services
GridDataService
GDS
Access to data & DB operations
GridDataServiceFactory
GDSF
Makes GDS & GDSF
GridDataServiceRegistry
GDSR
Discovery of GDS(F) & Data
GridDataTranslationService GDTS
Translates or Transforms Data
GridDataTransportDepot
Data transport with persistence
GDTD
Integrated Structured Data Transport
Relational & XML models supported
Role-based Authorisation
Binary structured files (later)
1a. Request to Registry
for sources of data
about “x”
SOAP/HTTP
Registry
1b. Registry
responds with
Factory handle
service creation
API interactions
2a. Request to Factory for access
to database
Factory
Client
2c. Factory returns
handle of GDS to
client
3a. Client queries GDS with
XPath, SQL, etc
3c. Results of query returned to
client as XML
2b. Factory creates
GridDataService to manage
access
Grid Data
Service
3b. GDS interacts
with database
XML /
Relationa
l
database
Interface transparency:
one GDS supports multiple
database types
Relational
database
Client
Client
Client
Grid
Data
Service
XML
databas
e
Director
y / File
system
Software Availability
Available now

Phase 1 prototype of GDS, GDSF
& GDSR for XML
Java implementations for the
axis/tomcat platform and the Xindice
database
•

Globus-2 Relational
database support
BinX Schema v0.2
www.epcc.ed.ac.uk/gridserve/WP5
An XML Schema for describing the
structure of binary datafiles – the
power of XML for terabyte files
Software Q1 2003

Reference implementation 1

Access & Update
•
•
XML databases
•
Relational databases
To be released as Basic
Services in Globus
Toolkit 3
umbriel.dcs.gla.ac.uk/NeSC/general/projects/OGSA_DAI/products
Advanced Components
Translation
Client
GDS:PerformScript
GDS
DB
Translation
GDT
Consumer
Composed Components
GDS:performScript
Translation
GDS:performScript
GDS
Client
GDS:performScript
GDT
Translation
GDS:performScript
GDT
GDT
Consumer
Futures of OGSA-DAI
Allow querying of distributed databases – this is using
Grid to federate multiple databases
Grid is “intrinsically” federation technology – need to mimic
classic database federation ideas in a Grid language
Form composite Schema from integration of those of individual
databases (OGSA-DAI allows you to query each database
web service to find schema)
Decide how to deal with very important case where
user view is a complex filter run on database query
Hardest when need to dynamically assign resource to perform
filter
Could view as a “simulation Web Service” outside OGSA-DAI
WSDL
Of Filter
Filter
DB
Semantic Grid starts with the Semantic Web which
is a “dream” and a project of W3C
“The Semantic Web is an extension of the current Web in which
information is given a well-defined meaning, better enabling
computers and people to work in cooperation. It is the idea of
having data on the Web defined and linked in a way that it can be
used for more effective discovery, automation, integration and
reuse across various applications. The Web can reach its full
potential if it becomes a place where data can be processed by
automated tools as well as people”
From the W3C Semantic Web Activity statement
Digital Brilliance is phase transition coming from “collective effect” in the Grid Spin
Glass.
• The Hosting environment is the “Ether”
• The Resources are the Spins
• The forces are the meta-data linking resources
• Knowledge (The Higgs) will emerge when we get enough meta-data to force phase
transition
Resource Description Framework
Richer semantics
Semantic
Web
Classical
Web
OWL Web Ontology Language
“The World Wide Web as it is currently constituted
resembles a poorly mapped geography. Our insight into
the documents and capabilities available are based on
keyword searches, abetted by clever use of document
connectivity and usage patterns. The sheer mass of this
data is unmanageable without powerful tool support. In
order to map this terrain more precisely, computational
agents require machine-readable descriptions of the
content and capabilities of web accessible resources.
These descriptions must be in addition to the humanreadable versions of that information.
The OWL Guide
SW Tools
Good Tools for recording meta-data (OWL) but not so
advanced in looking at their implications
• Semantic Web requires a metadata-enabled Web
• Where will the metadata come from?
• How about from the linked rich resources of a
virtual organization?
• A Grid …….
Classical
Web
Classical
Grid
More computation
Grid is metadata based middleware
Astronomy Sky Survey
Data Grid
1. Portals and Workbenches
2.Knowledge &
Resource
Management
3. Metadata
View
Catalog
Analysis
Bulk Data
Analysis
Standard APIs and Protocols
Concept space
4.Grid
Security
Caching
Replication
Backup
Scheduling
Data
View
Information
5. Discovery
Metadata
delivery
Data
Discovery
Data
Delivery
Standard Metadata format, Data model, Wire format
6.
Catalog Mediator
Data mediator
Catalog/Image Specific Access
7. Compute Resources
Derived Collections
Catalogs
Data Archives
An Example of RDF and Dublin Core
• <rdf:RDF xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
xmlns:dc="http://purl.org/metadata/dublin_core#">
• <rdf:Description about="http://www.dlib.org">
• <dc:Title>D-Lib Program - Research in Digital Libraries</dc:Title>
<dc:Description>The D-Lib program supports the community of people with
research interests in digital libraries and electronic publishing.
</dc:Description>
<dc:Publisher>Corporation For National Research Initiatives</dc:Publisher>
<dc:Date>1995-01-07</dc:Date>
• <dc:Subject>
– <rdf:Bag> <rdf:li>Research; statistical methods</rdf:li>
<rdf:li>Education, research, related topics</rdf:li>
<rdf:li>Library use Studies</rdf:li> </rdf:Bag>
• </dc:Subject> <dc:Type>World Wide Web Home Page</dc:Type>
• <dc:Format>text/html</dc:Format>
• <dc:Language>en</dc:Language>
• </rdf:Description> </rdf:RDF>
For example…
• Annotations of results, workflows and database entries could be
represented by RDF graphs using controlled vocabularies
described in RDF Schema and OWL
• Personal notes can be XML documents annotated with metadata
or RDF graphs linked to results or experimental plans
• Exporting results as RDF makes them available to be reasoned
over
• RDF graphs can be the “glue” that associates all the
components (literature, notes, code, databases, intermediate
results, sketches, images, workflows, the person doing the
experiment, the lab they are in, the final paper)
• The provenance trails that keep a record of how a collection of
services were orchestrated so they can be replicated or replayed,
or act as evidence
More meta-data …
• Represent the syntactic data types of e-Science objects using
XML Schema data types
• Represent domain ontologies for the semantic mediation
between database schema, an application’s inputs and outputs,
and workflow work items
• Represent domain ontologies and rules for parameters of
machines or algorithms to reason over allowed configurations
• Use reasoning over execution plans, workflows and other
combinations of services to ensure the semantic validity of the
composition
• Use RDF as a common data model for merging results drawn
from different resources or instruments
• Capture the structure of messages that are exchanged between
components
And more meta-data …
• At the data/computation layer: classification of computational
and data resources, performance metrics, job control,
management of physical and logical resources
• At the information layer: schema integration, workflow
descriptions, provenance trail
• At the knowledge layer: problem solving selection, intelligent
portals
• Governance of the Grid, for example access rights to
databases, personal profiles and security groupings
• Charging infrastructure, computational economy, support for
negotiation; e.g. through auction model
Richer semantics
http://www.semanticgrid.org
Semantic
Web
Semantic
Grid
Classical
Web
Classical
Grid
More computation
Source: Norman Paton
Summary of Grid Types
• Compute/File Grid: The “Linux workstation view of
distributed system” – need planning, scheduling of
10,000’s jobs, efficient movement of data to
processors
• Desktop Grid: as above but use huge numbers of
“foreign” compute resources
• Information Grids: Web service access to meta-data
rich data repositories
• Hybrid (complexity) Grids: Combination of Information
and Compute/File Grids
• Peer-to-peer Grid: Unstructured general purpose
access to other style grids
• Semantic Grid: Enables knowledge discovery in all
Grids