How to Accurately Measure Real‑Time Audio/Video Performance on Mobile Devices

Common Performance Metrics

For real‑time audio/video applications, transmission latency, clarity, feature completeness, and stability are primary concerns, but performance—especially on mobile devices—is also crucial for heat, standby time, and overall user experience.

Key mobile performance metrics include:

CPU usage

Memory usage

GPU usage

Display FPS and stutter rate

Network usage

Storage read/write

Battery consumption (power, battery level)

Temperature (CPU and battery)

NetEase Cloud IM’s NERTC SDK aims to deliver advanced functionality while minimizing system resource consumption.

Common Performance Analysis Tools

macOS/iOS tools

Xcode Debug Navigator : Shows CPU percentage, memory usage, performance impact, storage I/O, network traffic, and screen rendering FPS.

Instruments : Provides detailed CPU time, memory allocation/leaks, GPU performance, network, and energy logs. Components include Activity Monitor, Allocations, Leaks, Time Profile, Energy Log, Metal System Trace, and Network.

MetricKit : Offers XCTest Metrics, runtime power/CPU/network/memory statistics, and Xcode Metrics Organizer for visualizing battery performance.

Power Log : Offline system that records module‑level power data (voltage, current) at 20‑second intervals.

Android tools

Android Studio Profiler : Integrated suite for CPU, memory, network, and energy analysis.

Systrace / Perfetto : Precise timing data to understand device‑wide operations.

Simpleperf : Generates flame graphs to identify functions with high CPU consumption.

Third‑party tools

Several third‑party solutions can capture core metrics such as CPU/GPU usage, memory consumption, network traffic, rendering frame rate, and battery drain. NetEase also provides its own performance analysis tool to simplify testing.

Limitations of CPU Usage Percentage

CPU usage percentage is a common metric but ignores CPU frequency scaling, leading to inaccurate assessments for audio/video workloads. Experiments on iPhone devices showed that hardware encoding, which should reduce CPU load, reported higher CPU time than software encoding because the system lowered the CPU frequency.

Therefore, CPU usage percentage and time are unsuitable as sole performance indicators for audio/video scenarios.

Selecting Appropriate Metrics

For Android, the most accurate measurements require a rooted device with CPU frequency locked and only performance cores enabled.

For iOS and other platforms where frequency locking is unavailable, overall power consumption should be the primary metric, with CPU/GPU usage as auxiliary references.

How to Obtain Power Consumption Data

iOS Power Log provides detailed module‑level power information, and Battery voltage/current data can be used to calculate total device power draw. Android devices can also record current and voltage via code or third‑party tools to compute power consumption.

Evaluation Methodology

To ensure reliable performance evaluation, control the following variables:

Use a fixed test device (same model, optionally rooted for Android).

Maintain consistent environmental conditions (temperature, cooling).

Standardize device settings: same battery level, normal power mode, fixed screen brightness, disabled background tasks, turned‑off GPS/Bluetooth/NFC, identical speaker volume, Wi‑Fi conditions, and battery‑only power.

Use identical parameter configurations (codec resolution, frame rate, camera selection, etc.).

Run real‑time audio/video functions for a sustained period (30–60 minutes), then collect the average power consumption and auxiliary metrics. If power consumption deviates significantly from targets, developers should use debugging tools to pinpoint the cause.

Conclusion

Real‑time audio/video is a heavy‑load scenario where performance evaluation is essential yet challenging. Power consumption aligns closely with overall system overhead and heat generation, making it an effective holistic metric. NetEase Cloud IM combines video quality assessment with power‑based performance evaluation to deliver both excellent functionality and efficient performance.