Replicated Data Management using SRB

1
Replicated Data ManagementusingSRB
Arcot (Raja) Rajasekar San Diego Supercomputer
Center sekar_at_sdsc.edu
2
What is SRB?

• The Storage Resource Broker is a Middleware
• It provides uniform access to data in
  heterogeneous resources
• It uses a MetaCatalog to facilitate the brokering

MCAT
Application
SRB Server
HRM DB2, Oracle, Illustra, ObjectStore
HPSS, ADSM, UniTree
UNIX, NTFS, HTTP, FTP
3
SRB Concepts

• Abstraction of User Space
• Single sign-on
• Multiple authentication schemes
• certificates, passwords, tickets, group
  permissions, roles
• Virtualization of Resources
• Resource Location, Type Access transperancy
• Logical Resource Definitions - bundling
• Abstraction of Data and Collections
• Virtual Collections Persistent Identifier and
  Logical Name Space
• Replication Segmentation
• Data Discovery
• User-defined Metadata
• Attribute-based Access (path names become
  irrelevant)
• Uniform Access Methods
• APIs, Command Line, GUI Browsers, Web-Access
  (WSDL, CGI)
• Parallel Access with both Client and
  Server-driven strategies

4
The SRB Process Model
Application
Session Established
Authentication
1
Identification Initialization
(Host, port)
SRB Master
(port)
2
4
Server spawned
3
SRB agents
MCAT
5
Federated SRB Operation
Peer-to-peer Brokering
Read Application
Parallel Data Access
Logical Name Or Attribute Condition
5/6
6
SRB server
5/6
1
SRB server
3
4
5
SRB agent
SRB agent
2
Server(s) Spawning
R1
MCAT
1.Logical-to-Physical mapping 2. Identification
of Replicas 3.Access Audit Control
Data Access
R2
6
SRB Space
SRB
SRB
SRB
SRB
SRB
SRB
SRB
DL
DR - Data Repository DL - Dig Library MC - Meta
Catalog CP - Comp Process/ SRB Client
SRB
SRB
SRB
7
MCAT Metadata Catalog

• Stores metadata about
• Data sets, Collections, Users, Resources, Proxy
  Methods
• Maintains replica information for data
  containers
• Provides Collection abstraction for data
• Provides Global User name space
  authentication
• Provides Authorization through ACL tickets
• Maintains Audit trail on data collections
• Maintains metadata for methods and resources
• Provides Resource Transparency - logical
  resources
• Implemented as a relational database
• Oracle or DB2 or Sybase

8
SRB Replication Concepts

• Replication is a core functionality in SRB
• Logical Name Space (collection-hierarchies)
• local name (pathname/schema) independence
• replica can reside in any type of resource
• Automatic replica identification
• Persistent Id data movement independence
• Access Control at Replica Level
• Resource-level Access Control
• Replicas created using SRB or from outside
• Server-driven or client-driven replication
• Fault-tolerance, load balancing, caching
  archiving,
• Consistency Maintenance Strategies
• Data Access from Multiple Replicas (striping!)

9
SRB Replication Concepts (contd.)

• Multiple Replication Strategies
• Simple Vanilla (SV) Replication
• Logical Resource (LR) Replication
• Bundled Replication - using containers
• Semantic Replication
• Object Replicas
• Meta Information Replicas
• Inter-SRB Replication Multi-Grid
• BackUps (incremental)
• Mirroring collection-replication
• Data Placement Replication
• Cache management or load balancing based on
  replication strategies

10
SV Replication

• User-driven
• User chooses the resources to replicate
• In-Band Replication
• srbObjReplicate API
• Sreplicate command
• GUI Browser or the web
• Out-of-Band Replication
• Copy using non-SRB methods
• Register the copies as true replicas
• SRB does not check for consistency
• srbRegisterReplica API
• Consistency Maintenance
• Asynchronous -- modify one copy synchronize to
  others
• Synchronous modify all copies at once
• Locks

11
LR Replication

• User driven but server controlled
• Logical Resource LR PR1PR2
• Resource Typing cache, archive, permanent,
  volatile,
• Synchronous Replication
• Replication via Logical Resource definition
• Integrated into open/create write function
• Can choose k out of n asynchronous consistency
• Consistency Maintenance
• Asynchronous -- modify k/n synchronize later
  Ssyncd command
• Synchronous modify all copies at once
• Locks dirty bits
• Fail-safe Transaction Capability
• Automatic Cache Management Data Placements
• Mirroring (synchronous asynchronous)

12
Bundled Replication

• Use of Containers
• Small files are bundled into a container
• Data Movement enmasse
• SRB manages bundling individual file metadata
• Logical Resources associated with Containers
• Designated primary and secondary caches
• Writes go to primary cache
• Only one copy becomes writable
• Write Lock Protection
• Synchronize operation to archive copies
• Synch Purge Operation supported
• Fail-safe Transaction Capability
• Useful when data need to be co-located and used
  together
• Can associate containers with data collections
• All files in a collection go to same container

13
SRB Replica Reads

• Choice at Read
• any replica
• specific replica (by copy number)
• round-robin
• nearest
• by resource characteristics
• by timestamp or other characteristics
• data itself may be identified by meta
  charcteristics
• user defined metadata annotations
• data type, owner, comments, ...
• Fail-over Read tries a list of replicas in
  order
• Persistent Read retries until done

14
Semantic Replication

• Object Replicas
• Replica copies are not syntactically equal
• Example
• Jpeg, gif, tiff files of same image
• Word , pdf files of same document
• Same logical name and same persistent id
• Data Types and Data Sizes may vary
• Metadata Files a form of semantic replicas
• Synchronization of Metadata with Data changes
• Automatic Extraction of Metadata
• Triggers Background jobs

15
Multi-SRB Replication

• SRB-to-SRB replication
• Consistency Issues
• Each Copy is autonomous
• Locking across SRB space
• Write forced through Master-Slave Paradigm
• Meta-ownership issues
• New Issues with synchronous replication 3 phase
  commits
• Under Development

16
Conclusion

• SRB provides
• a uniform interface to heterogeneous data
  resources
• logical name space management - collections
• a replica management and mapping
• attribute-based data discovery access
• parallel access to replicated data
• SRB exposes multiple replica strategies
• Simple Vanilla (SV) Replication
• Logical Resource (LR) Replication
• Bundled Replication
• Semantic Replication
• Inter-SRB Replication Multi-Grid