Making video traffic a friendlier internet neighbor
Abstract/Contents
- Abstract
- Streaming video traffic from services like Netflix and YouTube accounts for the vast majority internet traffic today---recent estimates put the fraction as high as 60-75\% of all bytes sent over the internet. Given video traffic is such a large fraction of the internet, this thesis makes video traffic friendlier to neighboring applications that share the same networks. The thesis begins with Sammy, a system that smooths out video traffic so that the throughput of a video session is close to the minimum the video session needs for good performance. By judiciously picking throughput based on the needs of the video application, Sammy is able to substantially smooth out video traffic. In internet-scale experiments, Sammy reduces throughput of video traffic by more than half and dramatically reduce congestion, all while slightly improving video quality of experience over today's top production algorithms. The thesis next considers how new algorithms that affect congestion on the internet (like Sammy) are evaluated in networking research today. The gold standard is to run large-scale A/B tests, to understand at how an algorithm actually performs in practice. I show in experiments run at scale that typical A/B tests can lead to biased results, even to the point of switching the direction of results---an algorithm that appears worse in an A/B test might actually improve performance when deployed, and vice versa. I discuss the implications of this and suggest how researchers can deal with this bias. Finally, I revisit the long-standing problem of sizing buffers in routers. Prior work suggested that buffers can be reduced by a factor of square root of the number of flows using the router, offering dramatic buffer reductions in networks carrying many flows. I revisit these results, removing assumptions and showing that they hold for modern congestion control algorithms. Importantly, I discuss how our approach of smoothing video traffic with Sammy challenges the assumptions of the classic buffer sizing problem, potentially allowing for a future with much smaller buffers. Overall, this thesis invites us to reconsider how we deal with network congestion at scale. Rather than focusing on congestion control algorithms that maximize throughput, we should prioritize the ways we actually use the internet---applications like video streaming and web browsing and gaming and so on. If we do so, we can reduce congestion and improve the performance of all applications that share the internet.
Description
| Type of resource | text |
|---|---|
| Form | electronic resource; remote; computer; online resource |
| Extent | 1 online resource. |
| Place | California |
| Place | [Stanford, California] |
| Publisher | [Stanford University] |
| Copyright date | 2023; ©2023 |
| Publication date | 2023; 2023 |
| Issuance | monographic |
| Language | English |
Creators/Contributors
| Author | Spang, Bruce A |
|---|---|
| Degree supervisor | McKeown, Nick |
| Thesis advisor | McKeown, Nick |
| Thesis advisor | Johari, Ramesh, 1976- |
| Thesis advisor | Winstein, Keith |
| Degree committee member | Johari, Ramesh, 1976- |
| Degree committee member | Winstein, Keith |
| Associated with | Stanford University, School of Engineering |
| Associated with | Stanford University, Computer Science Department |
Subjects
| Genre | Theses |
|---|---|
| Genre | Text |
Bibliographic information
| Statement of responsibility | Bruce Spang. |
|---|---|
| Note | Submitted to the Computer Science Department. |
| Thesis | Thesis Ph.D. Stanford University 2023. |
| Location | https://purl.stanford.edu/tt015dr9566 |
Access conditions
- Copyright
- © 2023 by Bruce A Spang
- License
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...