Transcript Matt Denny - Berkeley Database Research

A Data Resolver Architecture for
Discovering Pervasive Data
Sources
Matthew Denny
Database Group
U.C. Berkeley
Where are the Data Sources in
Pervasive Applications?
• In traditional applications, the data sources are
well defined and reside at well-known locations
– SQL tables, web servers, SOAP/RPC apps, etc.
• In pervasive applications, neither property holds
– Data sources are not at any given location (cell phones
emitting diagnostic data roam about)
– Data sources may be unreliable (sensors may lose
power)
– Data sources that are used by one application may use
different protocols
Data Resolver Needed to
Discover Pervasive Data Sources
• Data Resolver allows applications to discover data sources
• Data Sources send advertisements to the data resolver
– Properties: name-value attribute pairs describing the data
– Interfaces: descriptions on how to access the data
• Applications send specifications to query the data resolver
– SQL or LDAP-like queries over the properties
• Application may want to know when data sources begin to
or no longer match the query
– Continuous Queries
– Subscriptions to a data source’s advertisements
Implementation Plan
• Utilize standards for queries and advertisements
– WSDL for service descriptions
• Scalability Problem: many rapidly updating data
sources
– Distributed “hybrid P2P” system with partial
replication
• Each DR Node caches data as specified by its Master DR Node
Specification
– Any node can accept any ad or query
• Publish-Subscribe system used to route ads
• Query Containment Indexing (derived from predicate indexing
research) used to route specifications