Roadmap

Development notes for pending tasks.

• Improved read buffer algorithm
  • Switch from PTL to Fast Flow based segments
• Dynamically grow read buffer segments
  • See Java 8's Striped64 for example heuristic
  • Reduces clean up penalty in low contention usages
  • Reduces memory usage
• Use relaxed reads and writes
  • Reduces CPU stalls due to memory barriers
  • Reduces memory usage by replacing atomics with volatile fields
• Improve write buffer algorithm
  • Use a multiple producer / single consumer unbounded queue (replacing ConcurrentLinkedQueue)
• Collections
  • Add combining arena to existing implementations
  • Investigate mpsc based on linked, reusable arrays
• Profile and code audit
  • Request external reviews
  • Implement atomic merge operation
  • Signal that incomplete async computations cannot expire
• Advanced eviction policies
  • Simulator implementations (2Q, LIRS, TinyLFU, etc)
  • Evaluate and adopt