Testing Curvy RED: to keep Internet Delays Low
As Internet capacity has increased, delay has become the critical factor limiting performance. Active Queue Management (AQM) is an important class of techniques for reducing queuing delay. The idea is to drop a few packets while the queue is shallow to signal to the end systems to stop increasing the load, so that they do not continually fill the buffer (part of a problem called buffer-bloat). In another project, we accidentally invented a new AQM algorithm that is extremely simple and appears to be very effective - it is also backed by solid theory, whereas other approach are not. We call it Curvy RED, because it is based on the original classic AQM algorithm random early detection (RED), but its packet dropping level rises as a power of the queue length, not just linearly.
We have already implemented Curvy RED in Linux and extensively tested some aspects, but only as a side-effect of the project from which it was a spin-off- this constrained Curvy RED too much. The goal of this project is to remove the constraints and test Curvy RED in its own right. A stretch goal is to design and test a better algorithm to control its dynamic behaviour. Depending on the parameter space, a decision will be needed on whether to test using network simulation or a testbed (or both).
- Traffic simulator and/or testbed techniques
- Internet traffic and congestion control techniques and AQM
- Linux kernel programming
- Writing an excellent research paper
- Methodical research approach
- Good simulation skills
- Good programming and scripting skills
- Good grounding in Internet protocols
- Good grounding in operating systems
Bob Briscoe, email@example.com