What is it about?
Drift-Bottle is a lightweight and distributed mechanism for real-time network monitoring. To detect and localize potential failures in the network, we consider the normal packets as drift-bottles which carry inferences about failures from one monitor to another to aggregate the information from all monitors.
Featured Image
Photo by Lakshya Thakur on Unsplash
Why is it important?
To reduce the overhead brought by monitoring system, we design a distributed mechanism which deploys its functions mainly on the switches, and it does not introduce an extra server for data collection and analysis. Drift-Bottle perform inference aggregation and analysis entirely in the switches and does not create any new packets. Thus, our system can detect failures without introducing considerable overhead to the network.
Perspectives
My favorite part of this paper is the distributed inference aggregation mechanism, which inserts the 'letter' of failure inference into the 'drift bottle' (a special header of normal packets) and lets the one who receives the drift bottle read and update the letter. I think the idea can be extended to other systems for lightweight network monitoring.
Xudong Zuo
Tsinghua University
Read the Original
This page is a summary of: Drift-bottle, November 2022, ACM (Association for Computing Machinery),
DOI: 10.1145/3555050.3569137.
You can read the full text:
Contributors
The following have contributed to this page
