Aurojit PandaInteresting (and scary) to see a blog post (https://thenewstack.io/research-ebpf-not-always-a-silver-bullet-for-network-apps/) about a new paper (led by @fshahinfar1.bsky.social, along with @gianniantichi.bsky.social and Sebastiano) explainging that the new shiny technology is not a panacea.
Paul Chaignon@apanda I was just telling @fshahinfar1 last week that it's not clear enough that the findings are fairly specific to application offloading into eBPF (which isn't a typical eBPF use case today). Seems it wasn't clear to the author of this article either 😕
@pchaigno @fshahinfar1 I agree with you: eBPF is a tool like anything else, and the findings are only about whether it makes sense to offload application functionality. There are many other networking use cases, including ones you have worked on @pchaigno for which it is a great fit.
@pchaigno @fshahinfar1 I'd go so far as to say that there are probably applications where offload makes sense.
@apanda @fshahinfar1 I guess my issue is that the term "networking application" is unclear and never defined. In the paper, the examples are offload from userspace, typically for L7 processing. But it's not very explicit and that nuance mostly disappeared in the NewStack article 😞
@pchaigno @fshahinfar1 That is fair: though none of us even knew about the NewStack article until yesterday, so we couldn't really have clarified the nuance.
@apanda @fshahinfar1 Yeah, I noticed they didn't even bother to contact you but did contact the eBPF Foundation for comment 😂