Skip to content

Latest commit

 

History

History
93 lines (64 loc) · 3.1 KB

PERFORMANCE.md

File metadata and controls

93 lines (64 loc) · 3.1 KB

Microverse Performance Guide

This guide documents how Microverse keeps CPU, energy impact, and memory usage low while preserving real-time system metrics.

Goals

  • Keep average process CPU low during normal use.
  • Reduce Energy Impact by minimizing unnecessary wake-ups.
  • Keep resident memory stable and avoid allocation churn.
  • Preserve metric accuracy vs macOS Activity Monitor.

Measurement Methods

1) Activity Monitor (process-level)

Use Activity Monitor to validate real app behavior:

  • % CPU (CPU tab)
  • Energy Impact and 12 hr Power (Energy tab)
  • Memory and Idle Wake Ups (CPU/Memory tabs)

Filter by Microverse, let the app settle for at least 2-3 minutes, and test both:

  • Idle state (no active interaction)
  • Active state (open popover, switch tabs, interact with widgets/notch)

2) Built-in CLI Benchmarks

Microverse includes a release-mode benchmark executable:

make benchmark

It benchmarks:

  • SystemCore syscall latency (getCPUUsage, getMemoryInfo)
  • Sequential vs TaskGroup overhead for CPU+memory sampling
  • Publish-suppression helpers (setIfChanged, quantization)
  • ByteCountFormatter usage patterns

Current Performance Architecture

Accurate CPU Sampling

  • SystemCore.SystemMonitor now uses HOST_CPU_LOAD_INFO tick deltas.
  • CPU usage is computed from interval deltas, matching Activity Monitor style sampling.
  • Counter rollover is handled with 32-bit wrapping arithmetic.

Accurate Memory Pressure

  • Memory pressure uses kernel signal kern.memorystatus_vm_pressure_level.
  • Heuristic fallback remains for sysctl failure.
  • speculative_count is not double-counted.

Demand-Driven Monitoring Lifecycle

  • SystemMonitoringService polling is ref-counted via acquireClient() / releaseClient().
  • Views opt in with .systemMonitoringActive(...).
  • Polling stops when no visible surface requires CPU/memory metrics.

Publish Suppression and Change Tokens

  • Metrics are quantized before publish (CPU integer bucket; memory rounded fields).
  • @Published updates occur only when display-visible values change.
  • sampleID: UInt64 replaces timestamp-style change signaling.

Wake-Up Reduction

  • Poll loops use Task.sleep(for:tolerance:) across system/network/audio/battery/weather paths.
  • Removed hard periodic force-update loop in adaptive notch metric mode.
  • Added one-shot dwell retries for adaptive switching so deferred switches are not lost.

Quick Validation Checklist

  1. Build and run:
    • make install-debug
  2. Validate process-level behavior in Activity Monitor:
    • CPU decreases at idle
    • Energy Impact decreases at idle
    • Idle Wake Ups trends lower
  3. Run synthetic checks:
    • make benchmark
  4. Functional smoke tests:
    • CPU/Memory tabs update in near real time
    • Notch auto metric switches correctly after dwell windows
    • Desktop widget adaptive primary switching still updates correctly

Notes

  • Benchmark numbers are machine- and thermal-state-dependent.
  • Prefer release-mode measurements (-c release) for any timing conclusion.
  • Activity Monitor remains the source of truth for end-user process impact.