R2 began from a desire to learn the internals of the HFSC scheduler by writing it from scratch. It started with our decision to use HFSC at a startup I was working at. We were having troubles with a home grown scheduler and decided to stop investing time and effort on that – rightly so since our engineering team’s size was in single digits. And the question was which open source scheduler to use. Someone else in the team had used BSD’s HFSC before, so the decision was quick. I took on the task of porting it and making it work the way we wanted. I didn’t want to just port it blindly without understanding the internals, because I knew I would have to fix bugs in it :). There really is just ONE piece of documentation on hfsc, which is the original paper with densely packed theorems and math.
Years flew by, and my anxieties went away, only because we stopped using HFSC and built something custom as the startup grew. But that paper remained in my bag and I kept reading it in every flight, it still stays in my bag wherever I travel. And at some point I thought I understood enough of the paper that it was time to write the code myself, but in a new language. Go was the popular kid in town and I thought I will write it in Go. At the same time my brother was experimenting with this new kid in town Rust and writing blogs about his experiments, so I started looking at Rust also. And for whatever reason, I just liked Rust more than Go and dropped Go and switched to Rust. And at some point when I got enough understanding of Rust and I was starting to write the scheduler, I thought why just write a scheduler, why not write an entire packet forwarding system. By then I had worked on plenty of networking code bases big and small, good and bad, success and failures, proprietary and open. And I had seen good designs and bad designs from others, made good designs and bad designs myself, and I thought why not put all the learnings together and build something bigger.
Years flew by, and my anxieties went away, only because we stopped using HFSC and built something custom as the startup grew. But that paper remained in my bag and I kept reading it in every flight, it still stays in my bag wherever I travel. And at some point I thought I understood enough of the paper that it was time to write the code myself, but in a new language. Go was the popular kid in town and I thought I will write it in Go. At the same time my brother was experimenting with this new kid in town Rust and writing blogs about his experiments, so I started looking at Rust also. And for whatever reason, I just liked Rust more than Go and dropped Go and switched to Rust. And at some point when I got enough understanding of Rust and I was starting to write the scheduler, I thought why just write a scheduler, why not write an entire packet forwarding system. By then I had worked on plenty of networking code bases big and small, good and bad, success and failures, proprietary and open. And I had seen good designs and bad designs from others, made good designs and bad designs myself, and I thought why not put all the learnings together and build something bigger.
