System Architecture

• C++ Overview
  • Applications
  • Libraries
• Introduction to CRTP / Facade
  • Layout
  • Analysis
  • Performance
  • Changing an Underlying Type
  • POLARIS Component Development
  • Conclusion
• Advanced CRTP
  • Deriving Polaris objects from other Polaris objects
• Random Choice
  • Normal and Bivariate Normal
  • Consistent seeding and reproducability
• Threading Model
  • Intro
  • Patterns
  • Synchronizing Shared State Across Threads
    • Simple shared values: use std::atomic
    • Compound operations and containers: use a lock
    • The _lock type
    • Option 1: ScopedLock (preferred outside of core)
    • Option 2: LOCK / UNLOCK macros (core hot paths)
    • Why spin locks under the hood
    • When to revisit
    • Lock variables in core
  • Exception Handling
• Threading Performance and agents_per_block
  • Threads vs simulate time
  • The agents_per_block cliff
  • Visualising core_diag.json
    • Per-thread step vs wait time
    • Threading efficiency
  • Takeaways
• Core Diagnostics
  • The files
    • core_diag.json is always produced
    • Producing the per-iteration CSVs
  • Analysing the output
• Simulation Engine
  • Memory Allocation Blocks
  • Hierarchy
  • Debugging
• Simulation Time
• Units
  • Transition from Basic_Units to units
• Routing
• Memory Usage
  • Analyzing memory usage
• Adding a new app
• Model Outputs
  • MOE Results as HDF5 Outputs
    • Working with H5 in C++
    • Working H5 in Python
  • Skims as OMX Outputs
    • Working With OMX in C++
    • Working With OMX in Python