Echo - Deep Dive

Echo is a work-stealing scheduler providing high-performance task execution for Mountain’s ApplicationRunTime and other components requiring async task management.

Architecture 🏗️

Built on crossbeam-deque for lock-free work-stealing queues. Supports three priority levels (High, Normal, Low) for latency-sensitive operations.

Core Components

Module	Role
`Task`	Unit of work with priority, generic future support, execution context
`Queue`	Lock-free work-stealing queue using `crossbeam_deque::Worker` and `Stealer`
`Scheduler`	Master coordinator with fluent builder API and worker pool
`SchedulerBuilder`	Configures worker count and scheduler parameters

Design Principles

Work-stealing: Idle workers steal from busy workers’ queues for load balancing
Lock-free: All queue operations use atomic instructions
Structured concurrency: Supervised worker pool with graceful startup/shutdown
Priority-aware: High/Normal/Low queues interleaved by worker threads
Failure resilient: Task failure does not crash the worker pool

Performance ⚡

Metric	Value
Task submission latency	~0.05ms
Queue operations	~0.02ms (lock-free push/pop)
Work-stealing transfer	~0.1ms
Total task latency	~0.18ms + execution time
Single-threaded throughput	1M tasks/second
Memory per task	< 64 bytes
Context switch cost	~100 nanoseconds

Scales linearly with available CPU cores.

Integration with Mountain 🔗

Echo is integrated as Mountain’s primary scheduler:

Mountain ApplicationRunTime
  -> submits ActionEffects as Tasks to Echo
  -> Echo distributes to worker threads
  -> Workers resolve capabilities and execute
  -> Results returned to ApplicationRunTime

Effect execution overhead when integrated: < 1 microsecond.