Performance Statistics

Last updated: 2021-10-19 15:50:47

    This document analyzes TRTC’s performance in terms of audio/video quality, latency, smoothness, stability, CPU usage, memory usage, battery consumption, heating, and other key indicators in tests under normal and poor network conditions and in different application scenarios (one-to-one, one-to-many, etc.).

    Performance Under Normal and Poor Network Conditions

    Scenario

    Video call, interactive live streaming, and audio call

    Parameter configuration

    • Video call:
      ParameterValue
      Resolution368 x 640
      Bitrate400 Kbps
      Frame rate15
    • Interactive live streaming:
      ParameterValue
      Resolution720 x 1280
      Bitrate1200 Kbps
      Frame rate15

    Poor network tolerance test

    The TRTC SDK was tested for its tolerance to different bad network conditions.

    Note:

    For the metrics used to measure tolerance to poor network conditions, please see Appendix 1: Network Metrics.

    Audio MOS under poor network conditions

    TRTC can deliver relatively high-quality audio and low latency under poor network conditions.
    The table below lists TRTC’s performance and mean opinion score (MOS) under different poor network conditions.

    Client SDK Performance

    Tested devices

    Device Processor Memory
    Android device 1 Qualcomm Snapdragon 835 - 8 cores 6 GB
    Android device 2 Kirin 980 - 8 cores 8 GB
    iOS device 1 Apple A8 - 2 cores 3 GB
    iOS device 2 Apple A13 - 6 cores 4 GB

    Parameter configuration

    Parameter Value
    Resolution 240 x 320
    Bitrate 100 Kbps
    Frame rate 15

    Test scheme

    • Scenario: one-to-one, one-to-two, one-to-three, one-to-four, one-to-eight
    • Duration: 30 min for each scenario
    • Method: a Linux robot is used to simulate streaming in one-to-many scenarios. Each device is tested independently.

    Test result

    The TRTC SDK performs well in terms of CPU usage, memory usage, heating, and battery consumption. It uses a small amount of hardware resources but provides quality audio/video services.

    • App CPU usage:
    • App memory usage:
    • System CPU usage
    • System memory usage
    • Battery drain after 30 min:
    • Heat increase after 30 min:

    Appendix 1: Network Metrics

    MetricDescriptionExample
    Loss Packet loss rate 50%: for every 10 packets sent, 5 are lost.
    Delay Network delay 200: Data packets are delivered by the network 200 ms after they are sent by the SDK.
    Jitter Network jitter 300: Packet sending may be delayed 20 ms, 50 ms, 250 ms, 280 ms, or any value up to 300 ms. The average delay is 150 ms.

    Appendix 2: Performance Metrics Under Poor Network Conditions

    Performance MetricDescription
    MOS An important measure of the audio quality of telecommunication systems. MOS is generated by Spirent Nomad using the POLQA standard. The higher the score, the higher the audio quality.
    End-to-end latency The time from when audio is captured at the sender end to when it is played back at the recipient end
    Poor network tolerance test Spirent Nomad is used to score the SDK under different poor network conditions using the POLQA standard. Foreman video sequences are used to send data, and frame intervals are monitored at the recipient end. Data is collected at 30 points over a course of 10 min or longer. If there are perceptible abnormalities of 3 min at more than 3 data points, or the SDK is unavailable for a relatively long period of time, the SDK is considered intolerant of the network conditions.
    Note:

    Perceptual Objective Listening Quality Analysis (POLQA) is the ITU-T P.863 standard. It is a globally applicable standard used to score speech quality under different network conditions.

    Appendix 3: SDK Performance Indicators

    IndicatorDescription
    App CPU usage Android Non-normalized CPU usage of the app, which is the same as the results generated by Android Studio Profiler
    iOS CPU usage of the app, which is the same as the results generated by Xcode
    PerfDog usage = Xcode usage / Number of cores
    System CPU usage Android Non-normalized CPU usage of the device, which is the same as the results generated by Android Studio Profiler
    iOS CPU usage of the device, which is the same as the results generated by Xcode
    PerfDog usage = Xcode usage / number of cores
    Memory usage Android Proportional set size (PSS), which is the same as the results generated by Android Java API and Meminfo
    iOS Xcode memory, which is obtained via debug gauges
    Battery drain Decrease in battery percentage after 30 min (calculation starts the moment the battery percentage drops from 100% to 99%.)
    Heat increase Temperature is measured with a thermometer when the app is not launched. Then run the app for 30 min under different scenarios.
    Heat increase = Temperature after 30 min – Temperature when the app is not launched