End of support: After enabling our own algorithm (Indigo) and assisting four algorithms from other research groups in publishing at NSDI '18 (Copa and Vivace), ICML '19 (Aurora), and SIGCOMM '20 (TCP-TACK), we regret to announce that we will no longer maintain Pantheon's codebase, accept new schemes, or run new experiments. Please check out our new video streaming platform Puffer, which also welcomes novel congestion-control algorithms to test with video traffic.
Welcome to the Pantheon of Congestion Control: a community evaluation platform for academic research on congestion control. The Pantheon provides several services to the designers of congestion-control schemes:
- A curated collection of 17 working implementations of congestion-control schemes, each wrapped with a Python script so as to expose the same interface to the calling application. Each scheme is continuously verified to compile and run by a CI system. We welcome contributions by means of a pull request from the academic community; these will be evaluated on a regular basis.
- A testbed of measurement nodes on wired and cellular networks in the USA, Mexico, Brazil, Colombia, India, and China, and in cloud datacenters in the UK, USA, Japan, and Australia. Stanford pays for the cellular and wired service in these locations. We regularly run all the congestion-control schemes in different workloads (single flow, or with scripted cross traffic) across the network paths between the nodes. We host a searchable archive of results.
- A collection of network emulators (using the
command-line mahimahi emulator) that can be used to test
congestion-control schemes locally. These include:
- emulators calibrated to match some of the above network paths.
- emulators exhibiting pathological network behaviors suggested by colleagues at Google (token-bucket-based policers, severe ACK aggregation, small bottleneck buffer sizes).
Learn more about Pantheon in the FAQ, and our paper at USENIX ATC 2018 (Best Paper Award winner). Please also feel free to inspect the code and make a pull request to add a new scheme on GitHub.
Below are the 10 most recent results. You can find more results on
the
Find Results
page, or view a
performance summary
of each scheme.
-
Performance summary graph (all runs)
-
Performance summary graph (mean of all runs)
- Full report
- Performance data
- Raw logs
- Raw logs with UIDs