Performance analysis of adaptive streaming algorithms for a low-latency environment
(2022) EITM01 20221Department of Electrical and Information Technology
- Abstract
- Streaming video over the internet can face issues when met with poor and varying network conditions, which can be especially noticeable when streaming live video such as security footage or video calling. To handle this, there exists congestion control algorithms that monitor network conditions based on feedback from the receiver and adapt the video output. This thesis aims to implement and compare three different algorithms NADA, SCReAM and GCC as part of Axis' video streaming system and IoT camera, and to perform a performative analysis in both environments.
To perform the analysis, a set of realistic situations are simulated inside the same framework that is used in production, as well as implementing and testing out a Proof on... (More) - Streaming video over the internet can face issues when met with poor and varying network conditions, which can be especially noticeable when streaming live video such as security footage or video calling. To handle this, there exists congestion control algorithms that monitor network conditions based on feedback from the receiver and adapt the video output. This thesis aims to implement and compare three different algorithms NADA, SCReAM and GCC as part of Axis' video streaming system and IoT camera, and to perform a performative analysis in both environments.
To perform the analysis, a set of realistic situations are simulated inside the same framework that is used in production, as well as implementing and testing out a Proof on Concept directly on camera hardware. The algorithms are implemented into Axis own WebRTC system, and run inside a container system with networking tools to monitor performance. A similar networking performance test is run directly on camera hardware to test out how it behaves on real hardware.
This thesis concludes that the NADA algorithm is an ideal choice when there is not a lot of latency present, having great utilization of the available link. However, in the presence of non-constant network delay it, together with SCReAM fails to utilize the link and only GCC can maintain a proper sending rate; GCC is shown to be a very good general-purpose algorithm. SCReAM shows a constant lower utilization of the available link, matching its target use of mobile networking and with the best measure of round-trip time in low-latency tests it is a good fit for remote-controlled devices. (Less) - Popular Abstract
- Congestion, be it traffic from cars or from computers, is an undesirable phenomenon. It causes slowdowns in the network and reduces the quality of service, which can compromise performance. In order to avoid this, the network can be monitored in order to look for possible signs of congestion, and regulate the data. This is especially important for real-time media, such as a live stream, as it both sends large amounts of data and when congested can show visible lag. For real-time media, congestion control algorithms can monitor the congestion present in the network, and adjust how much data may travel through it.
This thesis aims to test out how viable congestion control algorithms are inside WebRTC, a web standard for real-time... (More) - Congestion, be it traffic from cars or from computers, is an undesirable phenomenon. It causes slowdowns in the network and reduces the quality of service, which can compromise performance. In order to avoid this, the network can be monitored in order to look for possible signs of congestion, and regulate the data. This is especially important for real-time media, such as a live stream, as it both sends large amounts of data and when congested can show visible lag. For real-time media, congestion control algorithms can monitor the congestion present in the network, and adjust how much data may travel through it.
This thesis aims to test out how viable congestion control algorithms are inside WebRTC, a web standard for real-time communication, to perform dynamic changes to video quality. These tests are evaluated at Axis Communications, both as a simulation and on camera hardware.
In order to evaluate the performance of these congestion control algorithms, a simulation system is created inside Axis' systems to see how they perform in different network scenarios. These scenarios test a complete loss of network, a drop in available network and test where the latency is high. This system is used to evaluate the congestion algorithms called GCC, SCReAM and NADA. All three algorithms are used to monitor the network and estimate how high quality the video stream can be without causing congestion. These simulations are complemented with a test running on an Axis surveillance camera, to test the performance on real video hardware. For the camera, the NADA algorithm is tested with live video generated from the camera.
The results showed how GCC is able to keep a reliable video even in tests where high latency is introduced. NADA, however, is a well performing algorithm during changes in the network capacity when no additional latency is added, having a very fast response to these drops. The tests run very well on the camera, being able to properly react to changes in the network. However, this thesis seeks only to perform an initial evaluation, and optimization is the next goal for congestion control in Axis' cameras. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9090908
- author
- Kevo, Antonio LU and Sjölund, Albert LU
- supervisor
- organization
- course
- EITM01 20221
- year
- 2022
- type
- H2 - Master's Degree (Two Years)
- subject
- report number
- LU/LTH-EIT 2022-872
- language
- English
- id
- 9090908
- date added to LUP
- 2022-06-21 10:40:25
- date last changed
- 2022-06-21 10:40:25
@misc{9090908,
abstract = {{Streaming video over the internet can face issues when met with poor and varying network conditions, which can be especially noticeable when streaming live video such as security footage or video calling. To handle this, there exists congestion control algorithms that monitor network conditions based on feedback from the receiver and adapt the video output. This thesis aims to implement and compare three different algorithms NADA, SCReAM and GCC as part of Axis' video streaming system and IoT camera, and to perform a performative analysis in both environments.
To perform the analysis, a set of realistic situations are simulated inside the same framework that is used in production, as well as implementing and testing out a Proof on Concept directly on camera hardware. The algorithms are implemented into Axis own WebRTC system, and run inside a container system with networking tools to monitor performance. A similar networking performance test is run directly on camera hardware to test out how it behaves on real hardware.
This thesis concludes that the NADA algorithm is an ideal choice when there is not a lot of latency present, having great utilization of the available link. However, in the presence of non-constant network delay it, together with SCReAM fails to utilize the link and only GCC can maintain a proper sending rate; GCC is shown to be a very good general-purpose algorithm. SCReAM shows a constant lower utilization of the available link, matching its target use of mobile networking and with the best measure of round-trip time in low-latency tests it is a good fit for remote-controlled devices.}},
author = {{Kevo, Antonio and Sjölund, Albert}},
language = {{eng}},
note = {{Student Paper}},
title = {{Performance analysis of adaptive streaming algorithms for a low-latency environment}},
year = {{2022}},
}