Главная
Study mode:
on
1
Introduction
2
Key Design Aspects
3
Deploying Swift
4
Conclusions
5
Questions
6
Remarks
7
What is the fundamental problem
8
Future work directions
9
Simplifying congestion control
10
Why is Swift better than DC TCP or GCN
11
How to deal with artificial delay inflation
12
What is the initial window
13
What about stale information
14
Event growth and decrease proportional to measured delay
15
Do you use Swift for congestion control
16
Is burning CPU to run congestion control and issue for RDMA
17
Scaling limits
18
Reverse path congestion
19
Alternatives to Swift
Description:
Explore a conference talk from SIGCOMM 2020 that delves into Swift, a novel congestion control mechanism designed for datacenter networks. Learn about the key design aspects of Swift, which utilizes delay as a simple and effective approach to manage network congestion. Discover how Swift compares to other congestion control algorithms like DC TCP and GCN, and understand its deployment process. Gain insights into the fundamental problem Swift addresses, future work directions, and how it simplifies congestion control. Examine topics such as artificial delay inflation, initial window settings, handling stale information, and event growth proportional to measured delay. The talk also covers scaling limits, reverse path congestion, and potential alternatives to Swift. Engage with the Q&A session to deepen your understanding of Swift's implementation in RDMA environments, CPU usage concerns, and its overall impact on datacenter network performance.

SIGCOMM 2020 - Swift: Delay is Simple and Effective for Congestion Control in the Datacenter

Association for Computing Machinery (ACM)
Add to list
0:00 / 0:00