Video Streaming on Premium CPU-Optimized Droplets
Introduction
DigitalOcean recently launched Premium CPU-Optimized Droplets, virtual machines that deliver reliable performance and high network throughput, making them ideal for CPU and network-intensive workloads.
Key features of the Premium-CPU Optimized Droplets include:
- Premium Intel CPUs: Latest 1st and 2nd generations of Intel® Xeon® CPUs (58% better single core performance)
- Dedicated Hyper-Threading for better parallel processing
- NVMe SSD storage for faster disk I/O (290% faster writes to Droplet storage)
- Higher network throughput with up to 10 Gbps of bandwidth(5x the Non-premium Droplets)
Workloads optimum for Premium Droplets
For businesses with data-intensive workloads, utilizing Premium CPU-Optimized Droplets can maximize app performance reliability while leveraging higher outbound data speeds for a faster and smoother user experience. These Droplets also make it easy to scale operations, particularly for ML and AI development and training. Some of the ideal workloads include,
- Media streaming
- Analytics, data warehousing & machine learning
- Network & CPU intensive workloads: IoT, web crawling, real-time communication, ad-serving
In this article, we will pick one of the use cases, “media streaming,” and analyze how a real-world video conferencing platform performs on the newly launched Droplets. We will also run a similar setup on the “Non-premium Droplets” to compare the performance.
Benchmarking a Video Streaming Application
The setup
The benchmark setup consists of
-
Jitsi server, a popular open-source video conferencing application that allows users to create and customize their video conferencing server to their specific needs.
-
Premium CPU-Optimized Droplet
spec: 96 GB Memory / 48 Intel vCPUs / AMS3 - Ubuntu 22.04 (LTS) x64 -
Non-Premium CPU-Optimized Droplet
spec: 96 GB Memory / 48 Intel vCPUs / FRA1 - Ubuntu 22.04 (LTS) x64 -
webrtcperf, a benchmarking tool to test webrtc-based applications. This tool is also used to simulate over 100 users per video session.
The tests were run on both the Premium and the Non-premium Droplets throughout a twenty-minute video conference call
Installation & configuration
-
DigitalOcean offers a 1-click installation of Jitsi Server through our Marketplace, making it easy for users to set up and get started with video conferencing.
-
Before accessing the Droplet, add a domain to the project. For example,
chat.heartbyte.iois required because most browsers require HTTPS to use the camera and microphone on a website, and it is impossible to enable HTTPS just with an IP. Follow this guide to configure theA recordfor the public IP of the Droplet. -
Configure the Jitsi server
# Access the Droplet ssh root@Droplet-ip # Run the setup script and follow the instructions on the console ./complete-jitsi-setup.sh
Upon successful configuration, one should be able to access and start the Jitsi Meet session as shown below,
Benchmarking
Jitsi is a WebRTC-based video conferencing platform, and webrtcperf is a WebRTC performance and quality evaluation tool.
- It validates audio/video quality and the client CPU/memory usage when multiple connections join the same WebRTC service.
- It can be executed from sources using the pre-built Docker image or with the executables generated for each platform.
- It allows for throttling the networking configuration, limiting the ingress/egress available bandwidth, the RTT, or the packet loss %.
# Using containerized webrtcperf tool to simulate over 100 clients
docker pull ghcr.io/vpalmisano/webrtcperf
# Edit the `--url` flag with the jitsi meet URL
docker run -it --rm \
-v /dev/shm:/dev/shm \
ghcr.io/vpalmisano/webrtcperf \
--url="https://chat.heartbyte.io/benchmark#config.prejoinPageEnabled=false" \
--show-page-log=false \
--sessions=1 \
--tabs-per-session=100
Over a hundred clients are connected to the session for over 20 minutes on the Jitsi server running on a Premium CPU-Optimized Droplet in the Amsterdam data center.
Same tests were also run on the Jitsi server on the Non-premium Droplet in the Frankfurt data center.
Benchmarking in action
webrtcperfhas the instrumentation to measure the application’s performance running on the Droplet- We are streaming over 100 Linus Torvalds’ TED talks to the configured Jitsi server for simulation purposes
Results
Several key metrics are used to measure the performance and quality of video streaming applications. These metrics include:
- Bitrate(Audio/Video): Bitrate refers to the amount of data transmitted per second and is measured in kilobits per second (kbps) or megabits per second (Mbps). Higher bitrates generally result in higher-quality video but also require more bandwidth and may be more prone to buffering and other performance issues
- Packet loss(Audio/Video) is a metric measuring the percentage of data packets lost in transit between two devices. A higher packet loss rate can indicate network congestion or connectivity issues, which can affect a machine’s performance
- Jitter is a metric that measures the variation in the time it takes for data packets to reach their destination. Jitter can be caused by network congestion, packet buffering, and other factors that can cause delays in packet transmission. Higher jitter values can lead to increased latency and reduced network performance
- System CPU and memory performance
Observations
As expected, the Premium CPU-optimized Droplets outperformed the Non-Premium Droplets,
- 34.1% better CPU performance
- 26.75% less System memory usage
- Bitrate
- Video: 130.5% improvement in video outbound bitrate
- Audio: 1.26% improvement in audio outbound bitrate
- Packet loss
- Video: 0% packet loss on the Premium Droplets while Non-premium experienced a 0.95% packet loss
- Audio: 0% packet loss on the Premium Droplets, while Non-premium experienced a 2.25% packet loss
- Jitter
- Video: 86.95% less jitter
- Audio: 60% less jitter
Note: The benchmarking results might vary depending on hardware configuration, software settings, testing methodologies, and the specific workloads. Therefore, it is crucial to interpret benchmarking results cautiously and consider the context and variables involved in each test.
Conclusion
In conclusion, this tutorial comprehensively evaluates the performance and scalability of Jitsi, an open-source video conferencing software on DigitalOcean’s Premium and Non-premium Droplets. The results show that the Premium CPU-optimized Droplets better fit the video streaming use case. Therefore, this benchmark analysis can help you leverage the capabilities of Premium CPU-Optimized Droplets for your video-streaming applications. Check out the benchmarking-jitsi guide to get your hands dirty.
Happy benchmarking!
References
- Blog: Premium CPU-Optimized Droplets
- Video: Spin up Premium CPU-Optimized Droplets
- Benchmarking guide
Stats for nerds
-- Premium Droplets -------------------------------------------------------------------
name count sum mean stddev 5p 95p min max
System CPU 1 52.36 0.00 52.36 52.36 52.36 52.36 %
System GPU 1 0.00 0.00 0.00 0.00 0.00 0.00 %
System Memory 1 17.03 0.00 17.03 17.03 17.03 17.03 %
CPU/page 1 71.04 71.04 0.00 71.04 71.04 71.04 71.04 %
Memory/page 1 495.56 495.56 0.00 495.56 495.56 495.56 495.56 MB
Pages 3 10 3 0 2 4 2 4
Errors 1 795 795 0 795 795 795 795
Warnings 1 1301 1301 0 1301 1301 1301 1301
Peer Connections 3 6 2 1 0 4 0 4
-- Inbound audio -----------------------------------------------------------------------------------
rate 176 2523.38 14.34 16.85 0.00 40.01 0.00 40.33 Kbps
lost 80 0.00 0.00 0.00 0.00 0.00 0.00 %
jitter 176 0.01 0.00 0.00 0.01 0.00 0.02 s
avgJitterBufferDelay 80 206.82 285.72 60.27 1059.06 52.52 1130.43 ms
-- Inbound video -----------------------------------------------------------------------------------
received 122 384.99 3.16 1.29 0.01 5.01 0.00 5.10 MB
rate 117 9574.92 81.84 73.30 0.00 203.37 0.00 204.31 Kbps
lost 72 0.00 0.00 0.00 0.00 0.00 0.00 %
jitter 122 0.01 0.01 0.00 0.02 0.00 0.03 s
avgJitterBufferDelay 72 98.33 60.17 34.98 271.75 30.06 313.34 ms
width 119 320 0 320 320 320 320 px
height 119 180 0 180 180 180 180 px
fps 42 25 0 25 26 24 27 fps
-- Outbound audio ----------------------------------------------------------------------------------
rate 4 165.04 41.26 0.24 40.89 41.50 40.89 41.50 Kbps
lost 4 0.00 0.00 0.00 0.00 0.00 0.00 %
roundTripTime 4 0.003 0.001 0.002 0.004 0.002 0.004 s
-- Outbound video ----------------------------------------------------------------------------------
sent 4 98.14 24.53 21.26 0.56 48.20 0.56 48.20 MB
rate 4 4168.47 1042.12 964.97 4.18 2104.68 4.18 2104.68 Kbps
lost 4 0.00 0.00 0.00 0.00 0.00 0.00 %
roundTripTime 4 0.003 0.001 0.002 0.004 0.002 0.004 s
qualityLimitResolutionChanges 4 162 40 25 0 63 0 63
qualityLimitationCpu 4 92 23 23 0 54 0 54 %
qualityLimitationBandwidth 4 0 0 0 0 0 0 0 %
sentMaxBitrate 4 7200.00 1800.00 1600.00 200.00 3400.00 200.00 3400.00 Kbps
width 4 800 480 320 1280 320 1280 px
height 4 450 270 180 720 180 720 px
fps 4 22 5 13 25 13 25 fps
pliCountReceived 4 101 57 11 166 11 166
-- Non-Premium Droplets -------------------------------------------------------------------
name count sum mean stddev 5p 95p min max
System CPU 1 79.48 0.00 79.48 79.48 79.48 79.48 %
System GPU 1 0.00 0.00 0.00 0.00 0.00 0.00 %
System Memory 1 23.25 0.00 23.25 23.25 23.25 23.25 %
CPU/page 1 77.49 77.49 0.00 77.49 77.49 77.49 77.49 %
Memory/page 1 516.25 516.25 0.00 516.25 516.25 516.25 516.25 MB
Pages 2 10 5 4 1 9 1 9
Errors 1 684 684 0 684 684 684 684
Warnings 1 1576 1576 0 1576 1576 1576 1576
Peer Connections 2 1 0 0 0 1 0 1
-- Inbound audio -----------------------------------------------------------------------------------
rate 4 172.38 43.10 1.73 41.54 46.03 41.54 46.03 Kbps
lost 4 0.81 0.83 0.27 2.25 0.27 2.25 %
jitter 6 0.03 0.01 0.02 0.05 0.02 0.05 s
avgJitterBufferDelay 4 496.07 13.69 474.57 512.30 474.57 512.30 ms
-- Inbound video -----------------------------------------------------------------------------------
received 5 3.24 0.65 0.33 0.00 0.86 0.00 0.86 MB
rate 2 204.19 102.10 2.56 99.54 104.65 99.54 104.65 Kbps
lost 2 0.72 0.23 0.50 0.95 0.50 0.95 %
jitter 5 0.13 0.06 0.03 0.23 0.03 0.23 s
avgJitterBufferDelay 2 50.43 11.21 39.22 61.64 39.22 61.64 ms
width 4 300 34 240 320 240 320 px
height 4 168 19 135 180 135 180 px
-- Outbound audio ----------------------------------------------------------------------------------
rate 1 40.98 40.98 0.00 40.98 40.98 40.98 40.98 Kbps
lost 1 0.00 0.00 0.00 0.00 0.00 0.00 %
roundTripTime 1 0.040 0.000 0.040 0.040 0.040 0.040 s
-- Outbound video ----------------------------------------------------------------------------------
sent 1 7.07 7.07 0.00 7.07 7.07 7.07 7.07 MB
rate 1 912.75 912.75 0.00 912.75 912.75 912.75 912.75 Kbps
lost 1 0.11 0.00 0.11 0.11 0.11 0.11 %
roundTripTime 1 0.036 0.000 0.036 0.036 0.036 0.036 s
qualityLimitResolutionChanges 1 63 63 0 63 63 63 63
qualityLimitationCpu 1 0 0 0 0 0 0 0 %
qualityLimitationBandwidth 1 5 5 0 5 5 5 5 %
sentMaxBitrate 1 3400.00 3400.00 0.00 3400.00 3400.00 3400.00 3400.00 Kbps
width 1 1280 0 1280 1280 1280 1280 px
height 1 720 0 720 720 720 720 px
fps 1 8 0 8 8 8 8 fps
pliCountReceived 1 242 0 242 242 242 242