Despite lots of discussion in academia and industry, there isn’t consensus about how big a router buffer should be. Originally, we had the rule of thumb that the buffer should equal the bandwidth-delay product, but many networks no longer follow this guideline.
Choosing a buffer size is inherently complicated: It depends on (at least) the congestion control algorithm, AQM (marking and dropping policies), load-balancing, traffic engineering and the applications generating the traffic. So far, there has been some measurement and theory, but little consensus.
The goal of this workshop is to improve our understanding of how to pick buffer sizes in different parts of the network (e.g. DC, WAN, last mile, WiFi, cellular), for different applications (video, HPC, Hadoop-like, financial) and for different congestion control algorithms. Ideally, we would like to establish, from a mix of theoretical and experimental results, guidelines that can be used in practice to size routers’ buffers, and provide guidance to semiconductor and equipment providers.
Important dates:
- Now: notify us if you plan to submit or perform an experiment
- Paper registration (submission site): October 15th at 11.59pm Pacific Time.
- Paper submission (submission site): October 22nd at 11.59pm Pacific Time.
- TPC decision: November 5th.
- Final version due: November 22nd.
- Workshop presentations: December 2nd and 3rd.
Thank you to our sponsors
