Integration and Insight Aren

1
Integration and Insight Arent Simple Enough
Laura Haas IBM Distinguished Engineer Director,
Computer Science Almaden Research Center
2
Insight and Integration for the Enterprise

• Enterprises want business insight
• Need to correlate information from diverse
  sources
• Need powerful analytics and views
• Need to deliver results where, when, how needed
• Not easy
• Consumers may not understand sources or how to
  correlate them
• Different consumers have radically different
  needs and abilities
• Analytics and application subtle
• Goals
• automatically create a searchable source of all
  enterprise information (understand and integrate)
  
• enable a variety of actionable search and
  analytic capabilities over it (analyze and
  deliver)

3
Insight and Integration for the Enterprise

• Understand and Integrate
• Find relevant data sources
• Ingest large volumes of information
• Structured, semi-structured and unstructured data
• Real-time data streams and static warehouses
• Automatically discover semantics, identify
  important objects
• Automatically cleanse and reconcile data
• Identify data relating to the same entities from
  different sources
• Analyze and Deliver
• Provide simple ways to request complex analytics
• Keyword search, faceted search, natural language,
  new user interactions?
• Leverage semantics, find what is really desired
• Provide results in a form appropriate to the
  requestor, when/where needed
• Reports or visualizations or objects or
• Feed mashups, traditional programs, users in
  mobile or traditional environments
• New analytics, often solution-specific

4
Back to the Future? Nonprocedural Data Access is
What We Need (More Of)

• Relational dbms provided nonprocedural access
• Relational data model and declarative query
  language
• Relational calculus and algebra
• Optimizer and efficient algorithms for execution
• What has changed today?
• Heterogeneous data in overlapping heterogeneous
  sources
• Many if not most applications need data from
  multiple sources
• Users arent technical gurus lack knowledge of
  sources and query languages
• Needs are for deeper analysis, not just
  integration
• The tools we have today are far from relational
  simplicity
• Multiple integration engines with different
  capabilities
• Integration specification depends on engine,
  rarely declarative
• Operations defined by engines capabilities, no
  formal model