Peter Kunszt CERN Peter.Kunsztcern.ch

1
EU DataGrid Data Management Services

• Peter KunsztCERNPeter.Kunszt_at_cern.ch
• EU DataGrid WP2 Manager

2
Talk Outline

• Introdution to EU DataGrid workpackage 2
• WP2 Service Design and Interactions
• Replication Services
• Spitfire
• Security
• Conclusions and outlook

WP2 Members Diana Bosio, James Casey, Akos
Frohner, Leanne Guy, Peter Kunszt, Erwin Laure,
Levi Lucio, Heinz Stockinger, Kurt Stockinger -
CERN Giuseppe Andronico, Federico DiCarlo, Andrea
Domenici, Flavia Donno, Livio Salconi
INFN William Bell, David Cameron, Gavin McCance,
Paul Millar, Caitriona Nicholson PPARC,
University of Glasgow Joni Hahkala, Niklas
Karlsson, Ville Nenonen, Mika Silander, Marko
Niinimäki Helsinki Institute of Physics Olle
Mulmo, Gian Luca Volpato Swedish Research
Council
3
The EU DataGrid Project

• 9.8 M Euros EU funding over 3 years, twice as
  much from partners
• 90 for middleware and applications (HEP, Earth
  Obs. and Bio Med.)
• Three year phased developments demos
  (2001-2003)
• 2nd annual project review successfully passed in
  Feb 2003!
• Total of 21 partners
• Research and Academic institutes as well as
  industrial companies
• Related projects and activities
• DataTAG (2002-2003)
• CrossGrid (2002-2004)
• GRIDSTART (2002-2004)
• Grace (2002-2004)

4
EU DataGrid Project Objectives

• DataGrid is a project funded by European Union
  whose objective is to exploit and build the next
  generation computing infrastructure providing
  intensive computation and analysis of shared
  large-scale databases.
• Enable data intensive sciences by providing world
  wide Grid test beds to large distributed
  scientific organisations ( Virtual
  Organisations, VO)
• Start ( Kick off ) Jan 1, 2001
  End Dec 31, 2003
• Applications/End Users Communities HEP, Earth
  Observation, Biology
• Specific Project Objetives
• Middleware for fabric grid management
• Large scale testbed
• Production quality demonstrations
• To collaborate with and complement other European
  and US projects
• Contribute to Open Standards and international
  bodies
• ( GGF, IndustryResearch forum)

5
DataGrid Main Partners

• CERN International (Switzerland/France)
• CNRS - France
• ESA/ESRIN International (Italy)
• INFN - Italy
• NIKHEF The Netherlands
• PPARC - UK

6
Assistant Partners

• Industrial Partners
• Datamat (Italy)
• IBM-UK (UK)
• CS-SI (France)

• Research and Academic Institutes
• CESNET (Czech Republic)
• Commissariat à l'énergie atomique (CEA) France
• Computer and Automation Research Institute, 
  Hungarian Academy of Sciences (MTA SZTAKI)
• Consiglio Nazionale delle Ricerche (Italy)
• Helsinki Institute of Physics Finland
• Institut de Fisica d'Altes Energies (IFAE) -
  Spain
• Istituto Trentino di Cultura (IRST) Italy
• Konrad-Zuse-Zentrum für Informationstechnik
  Berlin - Germany
• Royal Netherlands Meteorological Institute (KNMI)
• Ruprecht-Karls-Universität Heidelberg - Germany
• Stichting Academisch Rekencentrum Amsterdam
  (SARA) Netherlands
• Swedish Research Council - Sweden

7
Project Schedule

• Project started on 1/Jan/2001
• TestBed 0 (early 2001)
• International test bed 0 infrastructure deployed
• Globus 1 only - no EDG middleware
• Successful Project Review by EU March 2002
• TestBed 1 ( 2002 )
• Successful 2nd Project Review by EU February
  2003
• TestBed 2 (Now)
• Some complete re-writes of components. Builds on
  TestBed 1 experience.
• TestBed 3 (Oktober 2003)
• Project stops on 31/Dec/2003, maybe a couple of
  months extension to wrap up and document results
  (no additional funding)

8
EDG Highlights

• All EU deliverables (40, gt2000 pages) submitted
• in time for the review according to the contract
  technical annex
• First test bed delivered with real production
  demos
• All deliverables (code documents) available via
  www.edg.org
• http//cern.ch/eu-datagrid/Deliverables/default.ht
  m
• requirements, surveys, architecture, design,
  procedures, testbed analysis etc.
• Project re-orientation last year in August From
  RD Testbed to production grid

9
Working Areas

• The DataGrid project is divided in 12 Work
  Packages distributed in four Working Areas

Applications
Middleware
Testbed
Management
Infrastructure
10
Work Packages

• WP1 Work Load Management System
• WP2 Data Management
• WP3 Grid Monitoring / Grid Information Systems
• WP4 Fabric Management
• WP5 Storage Element
• WP6 Testbed and demonstrators
• WP7 Network Monitoring
• WP8 High Energy Physics Applications
• WP9 Earth Observation
• WP10 Biology
  
• WP11 Dissemination
• WP12 Management

11
Trying hard to have a real GRID..

• Testbed 0
• Grid technology was not mature enough
• Configuration and deployment issues
• Stability problems
• Obscure errors
• Project reorientation Stability, Stability,
  Stability TB 1
• TB1 revealed a set of design bugs in Globus
• GASS Cache issue fixed by Condor (rewritten)
• MyProxy issues could never be used
• MDS did not scale had to set up fake local info
  system
• Reingeneering of essential components TB 2
• New resource broker
• R-GMA instead of MDS as info system
• Concrete Support channels (VDT)
• New configuration tool LCFG-ng (from U of
  Edinburgh!)

12
Grid middleware architecture hourglass

• Current Grid architectural functional blocks

CMS
ATLAS
CMS
LHCb
Specific application layer
Earth Observation and Biomed
HEP Application Services (LCG)
Common application layer
EU DataGrid middleware
High Level Grid Services
Basic Grid Services
GLOBUS 2.2
OS, Storage Network services
13
EU DataGrid WP2Data Management Work Package

• Responsible for
• Transparent data location and secure access
• Wide-area replication
• Data access optimization
• Metadata access
• NOT responsible for (but it has to be done)
• Data storage (WP5)
• Proper Relational Database bindings (Spitfire)
• Remote I/O (GFAL)
• Security infrastructure (VOMS)

14
WP2 Service Paradigms

• Choice of technology
• Java-based servers using Web Services
• Tomcat, Oracle 9iAS, soon WebSphere
• Interface definitions in WSDL
• Client stubs for many languages (Java, C, C)
• Axis, gSOAP
• Persistent service data in Relational Databases
• MySQL, Oracle, soon DB2
• Modularity
• Modular service design for pluggability and
  extensibility
• No vendor specific lock-ins
• Evolvable
• Easy adaptation to evolving standards (OGSA, WSDL
  1.2)
• Largely independent of underlying OS, RDBMS
  works on Windows too!

15
Replication Services Basic Functionality
Each file has a unique Grid ID. Locations
corresponding to the GUID are kept in the Replica
Location Service.
Users may assign aliases to the GUIDs. These are
kept in the Replica Metadata Catalog.
Files have replicas stored at many Grid sites on
Storage Elements.
Replica Metadata Catalog
Replica Location Service
Replica Manager
The Replica Manager provides atomicity for file
operations, assuring consistency of SE and
catalog contents.
Storage Element
Storage Element
16
Higher Level Replication Services
The Replica Subscription Service issues
Replication commands automatically, based on a
set of subscription rules defined by the user.
Hooks for user-defined pre- and post-processing
for replication operations are available.
Replica Subscription Service
The Replica Manager may call on the Replica
Optimization service to find the best replica
among many based on network and SE monitoring.
Replica Metadata Catalog
Replica Location Service
Replica Manager
Replica Optimization Service
Storage Element
Storage Element
SE Monitor
Network Monitor
17
Interactions with other Grid components
Virtual Organization Membership Service
User Interface or Worker Node
Resource Broker
Replica Subscription Service
Information Service
Replica Metadata Catalog
Replica Location Service
Replica Manager
Replica Optimization Service
Applications and users interface to data through
the Replica Manager either directly or through
the Resource Broker. Management calls should
never go directly to the SE.
Storage Element
Storage Element
SE Monitor
Network Monitor
18
Replication Services Status

• Current Status
• All components are deployed right now except
  for the RSS
• Initial tests show that expected performance can
  be met
• Need proper testing in a real user environment
  EDG2 LCG1
• Features for next release
• Currently Worker Nodes need outbound connectivity
  Replica Manager Service needed. Needs proper
  security delegation mechanism.
• Logical collections support
• Service-level authorization
• GUI

19
Spitfire Grid-enabling RDBMS

• Capabilities
• Simple Grid enabled front end to any type of
  local or remote RDBMS through secure SOAP-RPC
• Sample generic RDBMS methods may easily be
  customized with little additional development,
  providing WSDL interfaces
• Browser integration
• GSI authentication
• Local authorization mechanism
• Status current version 2.1
• Used by EU DataGrid Earth Observation and
  Biomedical applications.
• Next Step OGSA-DAI interface

20
Spitfire added value Security

• Grid security
• TrustManager deals with GSI proxy certificates
• Support for VOMS certificate extensions
• Secure java, c/c, perl clients
• Local Authorization
• Mapping through Gridmap file supported
• Fine grained authorization hooks a mapping
  service is provided to map VOMS extensions
  (group, role, capability) to DB roles that
  depending on the DB may be row-level
  authorization mechanisms (GRANT/DENY).
• Installation kit
• Easy installation and configuration of all
  security options

21
Spitfire customization

• Spitfire started as a proof of technology for
  Web Services and java.
• Customizable into specific services dealing with
  persistent data
• All WP2 services are in this sense Spitfire
  services (see later)
• Test platform for latest available codebase
• Gained experience with WSDL, JNDI, Tomcat, Axis,
  gSOAP
• Next things to try JBOSS (for JMS, JMX)
• Experimental add-ons
• Secure browser using JSP (proxy certificates for
  mozilla, netscape, ie..)
• Distributed query agent drop-in
• Todo OGSA-DAI interface as far as possible

22
RLS Architecture (evolved!)
A hierarchical RLS topology LRCs update RLIs,
RLIs may forward information
RLI
RLI
RLI
RLI
RLI
RLI
RLI
RLI
LRC
LRC
LRC
LRC
LRC sending updates to all Tier 1 RLIs
RLIs indexing over the full namespace (all LRCs
are indexed) receiving updates directly
RLI receiving updates from other RLIs
23
EDG Grid Catalogs (1/2)

• Replica Location Service (RLS)
• Local Replica Catalog (LRC)
• Stores GUID to Physical File Name (PFN) mappings
• Stores attributes on PFNs
• Local Replica Catalogs in Grid One per Storage
  Element (per VO)
• Tested to 1.5M entries
• Replica Location Index (RLI)
• Allow fast lookup of which sites store GUID -gt
  PFN mappings for a given GUID
• Replica Location Indices in the Grid Normally
  one per Site (per VO), which indexes all LRCs in
  the Grid
• Being deployed as part of EDG 2.1 in July
• In the process of integration into other
  components
• Tested to 10M entries in an RLI

24
EDG Grid Catalogs (2/2)

• Replica Metadata Catalog (RMC)
• Stores Logical File Name (LFN) to GUID mappings
  user-defined aliases
• Stores attributes on LFNs and GUIDs
• One logical Replica Metadata Catalog in Grid (per
  VO)
• Single point of synchronization current
  assumption in EDG model
• bottleneck ? - move to replicated distributed
  database
• No Application Metadata Catalog provided see
  Spitfire
• But Replica Metadata Catalog has support for
  small level of application metadata O(10)
• RMC usage not as well understood as Replica
  Location Service
• Architectural changes likely
• Use cases required

25
Typical Location of Services in LCG-1
CERN
Storage Element
Replica Metadata Catalog
Replica Location Index
Local Replica Catalog
Replica Location Index
Local Replica Catalog
Storage Element
Storage Element
RAL
26
Catalog Implementation Details

• Catalogs implemented in Java as Web Services, and
  hosted in a J2EE application server
• Uses Tomcat4 or Oracle 9iAS for application
  server
• Uses Jakarta Axis for Web Services container
• Java and C client APIs currently provided using
  Jakarta Axis (Java) and gSoap (C)
• Catalog data stored in a Relational Database
• Runs with either Oracle 9i or MySQL
• Catalog APIs exposed as a Web Service using WSDL
• Easy to write a new client if we dont support
  your language right now
• Vendor neutral approach taken to allow different
  deployment options

27
Quality of Service

• Quality of Service depends upon both the server
  software and architecture used as well as the
  software components deployed on it
• Features required for high Quality of Service
• High Availability
• Manageability
• Monitoring
• Backup and Recovery with defined Service Level
  Agreements
• Approach
• Use vendor solutions for availability and
  manageability where available
• Use common IT-DB solutions for monitoring and
  recovery
• Components architected to allow easy deployment
  in high-availability environment
• A variety of solutions with different
  characteristics are possible

28
Tradeoffs in different solutions
Clustered Oracle 9i/9iAS
Clustered Oracle 9i/Tomcat
Availability
Single Instance MySQL/Tomcat
Single Instance Oracle 9i/9iAS
Manageability
29
Current Deployment Plans

• EDG
• All sites use MySQL/Tomcat single instance
  solution
• LCG-1
• CERN deploys LRC/RLI/RMC on Oracle 9iAS/Oracle 9i
  single instance
• Tier-1 sites invited to use either Oracle
  9iAS/Oracle or Tomcat4/MySQL single instance for
  their LRC/RLIs
• CERN IT-DB working on easy-install packaging of
  Oracle
• Oracle sees ease of install as a high priority
  for Oracle 10i
• release date - Nov 2003
• Allow deployment of an Oracle based solution
  without requiring a lot of Oracle expertise
• Testing of components for high-availability
  solution in progress
• Based on Oracle 9i
• Plan to be available for year-end 2003

30
System Architecture High Availability

• Standard n-tier architecture
• Front end application layer load-balancer
• Oracle 9iAS Web Cache
• Cluster of stateless application servers
• Oracle 9iAS J2EE container
• Clustered database nodes
• Oracle 9i/RAC
• Shared SAN storage
• Fibre Channel storage

31
Security Infrastructure for Java-based Web
Services

• Trust Manager
• Mutual client-server authentication using GSI (ie
  PKI X509 certificates) for all WP2 services
• Supports everything transported over SSL
• Authorization Manager
• Supports coarse grained authorization Mapping
  user-gtrole-gtattribute
• Fine grained authorization through policies, role
  and attribute maps
• Web-based Admin interface for managing the
  authorization policies and tables
• Status
• Fully implemented, authentication is enabled on
  the service level
• Delegation implementation needs to be finished
• Authorization needs more integration, waiting for
  deployment of VOMS

32
Conclusions and outlook

• Re-focus on production has been a good but
  painful choice
• from hype to understanding the implications of
  wanting to run a production Grid
• reengineering of several components was necessary
• however, the project was not well prepared for
  this change the timelines had to be constantly
  revised in the last year
• The second generation Data Management services
  have been designed and implemented based on the
  Web Service paradigm
• Flexible, extensible service framework
• Deployment choices robust, highly available
  commercial products supported (eg. Oracle) as
  well as open-source (MySQL, Tomcat)
• First experiences with these services show that
  their performance meets the expectations
• Real-life usage will show its strengths and
  weaknesses on the LCG-1 and EDG2.0 testbeds
  during the rest of this year.
• Proceed with standardization efforts DAI, RLS
• Carry over the experience into the next project
  EGEE