Ask HN: Looking for a fantastic old talk on kernel network abstractions | Hacker News Reader

Ask HN: Looking for a fantastic old talk on kernel network abstractions

Hi, it's probably been at least 10 years now since there was a talk about the design of network abstractions within an operating system kernel. I think it may have been by someone from the BSD camp, as opposed as to someone from Linux. The slides from the talk really elegantly addressed issues about the kind of devices and various complexities around abstracting driver interfaces whilst dealing with differing protocols, hardware interfaces, and other issues. I would really appreciate finding this talk again. Please help, I'm sorry I can't provide more details. I seem to have lost my old bookmark.

View on HN

To add some further recollection. The talk wasn't about what the code presently was, it was aspirational: about how the structure could be wholly rearchitectured. The goal was simplicity. In networking, you have protocols which a varied and can be mixed and matched and layered, devices which can be plain or have different and varied off-load capabilities, things like power management and hot-plugging, and yet in all this you want to architect code that is modular and has maximal re-use and yet is also high-performance. This talk predated things like wireguard (which presents as a device implemented in software) and network namespaces but you got the impression that it was feeling out these possibilities without explicitly articulating them.

Sounds very interesting. Would love to see it as well.

Let us know if you find it. Sounds really interesting.

Was it Van Jacobson's Netchannels? https://www.slideshare.net/susantsahani/van-jaconson-netchan...

That was a very influential talk, but no that was not what I was thinking. Thank you, however.

Some of this makes me think of the Exokernel architecture [1].

[1] https://en.m.wikipedia.org/wiki/Exokernel

Another possibility is Microsoft Research's Singularity Project [1]. I read about both the Exokernel project and the Singularity project around the same time, so may have ideas from both conflated into one.

[1] https://www.microsoft.com/en-us/research/project/singularity...

Perhaps this was one of the talks by Radia Perlman?

She's great but it wasn't her. I remember one of her keynotes at Linux.conf.au being really wonderful. If people haven't see that, it's: https://www.youtube.com/watch?v=2JMRfVxiTNU

Reply on news.ycombinator.com