Around mid-March, Bootlin engineers Köry Maincent And Romain Gantois We had the opportunity to take part in the 0x19 of the NetDEV conference, where developers of the Linux Kernel Networking Community gather technical presentations and discussions for four days.
They were generally very impressed by the quality and details of the discussions held during this conference and selected two of the conversations that found them most interesting that they summarized below. We expect the films and videos to be published at some point in the future, but they do not seem to be available at the moment.
Understanding and improving the robustness of networked systems with machine learning
Talk of Hellden Godfrey
This keynote by Brian Godfrey, professor at the UIUC, focused on possible applications of mechanical learning to understand and manage network systems. Pr. Godfrey pointed out that these systems were usually quite unpredictable after use, but they presented the advantage of slightly available power metrics. According to him, this combination of properties was the main reason for machine learning. To demonstrate this, he presented two examples in which machine learning was used to solve networked system management problems.
The first example was the troubleshooting of incidents in networked systems with distributed applications. In such systems, especially if the number of companies involved is quite high, it can be very difficult to close a basic cause from a sudden local reduction in services. Pr. Godfrey constructed a machine learning model called Murphy, which could be trained to learn a common distribution of probability for all power metrics in the entity relationship graphic. This model could then be used to create “explanatory diagrams”, which linked local changes to the metrics with probable basic causes in the graph of the entities.
Talk abstract on the NetDev Conference website
Challenges and limits of the debugging of increasingly complex virtualized networks with retis
Speech by Antoine Tenart, Paolo Valerio, Adrián Moreno
The meeting began to describe the challenges in the pursuit of packages through various levels, protocols and drivers in Linux network data aoph. This complexity complicates the development of functions and error removal processes, which makes it increasingly difficult to effectively diagnose and fix network problems.
In order to cope with these challenges, the moderators Retis introduced, a tool that uses the flexibility of EBPF and integrated into the LibPCAP filtering. Retis was developed to monitor packages of collecting context information from components such as NFTABLES, Conntrack and Open Vswitch and enabling the data collected.
The tutorial dealt with the architectural decisions behind Retis, shaped by the identified debugging challenges. Important features and implementation strategies from retis have been thoroughly examined, which conveyed a comprehensive understanding of his skills.
During the entire lecture, several examples were used to demonstrate how retis can be used to get insights into internal package processing. These practical scenarios have emphasized the effectiveness of the tool in the troubleshooting and development environments.
The session ended with a discussion about the current restrictions of retis and potential ways for future improvements to invite collaborative inputs to further refine the tool.
Overall, the tutorial offered valuable perspectives on the debugging of complex networks and emphasized Retis as a promising solution in this area. This tool certainly helps us to debug the network stack, to make more efficient network analysis and to remedy the troubleshooting.
Talk abstract on the NetDev Conference website
Diploma
Overall, participation in NetDEV 0x19 was a very enriching experience for our team. The deep and technical quality of the sessions increased the value of this conference for everyone who is involved in Linux -Kernel networks. We are particularly pleased to apply some of the knowledge gained in our future work. Many thanks to the NetDev Society for your efforts to organize this great event! We are already looking forward to the next edition.