Skip to content

Analytic Architecture Framework

  • "Architecture Cube" has 3 dimensions:

  • Level of definition: Reference, Logical, Solution, Deployment

    • Context: Business, Information, Application, System
    • Time: Current, Transition, Target Architectural Framework

  • UDA design UDA

Teradata Reference Information Architecture

  • Consists of 3 layers: Acquisition, Integrated and Access. Additionally has,
  • Data Lab to host un-integrated data for quick prototyping
    • metadata (name, definition, format and length)

Access Layer

  • Base Tables -> Core (1:1) views -> Access Control -> Semantic -> Users
  • Access Control: Security, Privacy and Bypass
  • A Relational Model focuses on capturing business rules
  • A Dimensional Model focuses on evaluation, that is, monitoring business through metrics
  • The relationships in a dimensional model represent navigational paths v/s business rules in relational model
  • Has measures, such as, amounts, counts, duration, that are mathematical
  • Has meters, that is fact tables, that servers as buckets for measures
    • Grain is meter’s lowest level of detail
  • Has dimensions, are various ways to aggregate/filter measures
  • Supports Navigations, such as drill up/down and across (cross meter measurements)
  • Normal red flags: normalized structures, fuzzy grain, subtypes, too abstract
  • Most Semantic Models are Dimensional Models, but can be
  • Relational or Dimensional
  • Conceptual, Logical or Physical
  • Solution Modeling Building Blocks are collection of data model, SQL Views and mappings
  • SMBB are to the access layer as iLDM are to Integrated Data Layer
  • Building Steps:
  • Plan: Know the: deliverables, timelines, cultures+biases, skills+resourcs
  • Elicit Requirements: Interviews, Workshops, Documents, Prototyping
  • Scope: Agree on business questions that will be answered.
    • Grain Matrix tool: captures level of reporting for each measure
    • Conceptual Data Model: defines business vocabulary
  • Design: Build business solution (LDM) and technical solution (PDM)
    • LDM: Add measures and dimension attributes to CDM
    • PDM: Decide between star schema and snowflake
    • physical implementation (views > PDM extensions (JI) > Denorm)