BBRv2 aims to solve a different issue -- which is that in shallow-buffered networks, BBR consumes the entire link and starves out Cubic and Reno completely.
This work shows something different. Even in deep-buffered networks (like over most Internet access links) BBR(v1) has a funny property that it takes a fixed fraction fo the network link -- no matter how many other competing TCP connections there are. The deeper the buffer, the closer this fraction edges towards 50%.
Given that there are a lot of changes from BBRv1 to BBRv2, it's not at all clear whether this newer problem will exist in BBRv2 (we're trying to run some experiments to find out soon!)
Nefeli Networks | Software Engineer, SRE | Berkeley, California | Onsite, Full Time
We're a stealth-mode, early-stage networking startup spun out of UC Berkeley looking for Software Engineers and SREs. We have a ton of hard problems on high-performance software data planes and complex distributed NFV orchestration. Our systems are primarily in C++ with some Python, Rust, and Go.
The University of Waterloo Computer Engineering department offers a "Co-Op" Engineering degree. My understanding is that the students take five years to graduate, rather than the traditional four, and alternate by semester between work and school. One semester in classes, the other semester at internship. The graduating students I've met absolutely dominate technical interviews and the like.
The Rochester Institute of Technology has also has Co-Ops as a graduation requirement for most technical and engineering degrees. The number of quarters required depended on the degree. For example, I had to do four quarters for my computer science degree.
It was a fantastic experience, and every job I've held has come to me in some way through contacts I made through that program. Plus, it was extremely valuable to come back after a co-op and connect what you were learning with the real world.
The school I'm attending now, University of Cincinnati, has a compulsory co-op program for the majority of technical degrees. I'm going into my last year of school having around a year and a half of work experience already. I've learned just as many things while working as I have from taking classes.
It's actually all Engineering departments. The Science, Arts and Math faculties also have either optional or compulsory coop degrees for their students.
BBRv2 aims to solve a different issue -- which is that in shallow-buffered networks, BBR consumes the entire link and starves out Cubic and Reno completely.
This work shows something different. Even in deep-buffered networks (like over most Internet access links) BBR(v1) has a funny property that it takes a fixed fraction fo the network link -- no matter how many other competing TCP connections there are. The deeper the buffer, the closer this fraction edges towards 50%.
Given that there are a lot of changes from BBRv1 to BBRv2, it's not at all clear whether this newer problem will exist in BBRv2 (we're trying to run some experiments to find out soon!)