Real-world network throughputs traces used in MERINA(+) [paper].
Nuowen Kan, Yuankun Jiang, Chenglin Li, Wenrui Dai, Junni Zou, and Hongkai Xiong. 2022. Improving Generalization for Neural Adaptive Video Streaming via Meta Reinforcement Learning. In Proceedings of the 30th ACM International Conference on Multimedia (MM '22). Association for Computing Machinery, New York, NY, USA, 3006–3016. https://doi.org/10.1145/3503161.3548331
A collection of four public real-world network throughput datasets for simulating various user and network conditions. All traces are pre-processed with the format of [time_stamp (sec), throughput (Mbit/sec)]. Refer to Pensieve for more details.
-
3G/HSDPA dataset, published in Commute Path Bandwidth Traces from 3G Networks, which is stored in
./cooked_3gp/and./traces_3gp. -
FCC dataset, collected from Raw Data-Measuring Broadband America. (2016), which is stored in
./fcc_ori/. -
FCC & HSDPA dataset, which includes mixed traces from FCC and 3G/HSDPA and is stored in
./fcc_and_hsdpa/. -
Oboe dataset, published in Oboe: auto-tuning video ABR algorithms to network conditions, which is stored in
./traces_oboe/. -
Puffer dataset, collected from Puffer platform. We precess all traces on two randomly chosen dates (Oct. 17, 2021 and Feb. 18, 2022), and the results are stored in
./puffer_211017/and./puffer_220218/, respectively. -
./load_webget_data.pyshows how to process raw Puffer traces.