Bridging the Gap between QoE and QoS in Congestion Control: A Large-scale Mobile...
summary
Builds Floo, a shim layer between apps and CCAs that monitors app state and network metrics, runs a reinforcement learning selection policy, and performs fast state and variable migration to switch CCAs without packet loss. Deployed to 5% of a 10 million DAU mobile web service, collects 35 million requests, and shows end-to-end request completion time improvements (average ~14% and up to 25% on the 99th percentile) and an additional ~7% gain from state migration. CCA switching at runtime
AI-written summary · about
chapters (9)
- 0:00Intro
- 4:23Design
- 8:53Challenges
- 10:22Phase Migration
- 11:54Variable Migration
- 13:39Experimental Results
- 14:22Real World Results
- 15:09Transport Layer Performance
- 16:37Conclusion
related talks
16mQCON: Seamless QoE-Aware 5G Streaming via Multi-Connectivity
15mMortise: Auto-tuning Congestion Control to Optimize QoE via Network-Aware Parameter
18mCCC: Re-architecting Delay-based Congestion Control in Datacenter Networks
19mDChannel: Accelerating Mobile Applications With Parallel High-bandwidth...
14mA large-scale deployment of DCTCP
16mBuilding A CSFQ-Inspired Transport for Switched CXL Memory Pooling
43mHow internet congestion control actually works in the bufferbloat age
19mLearning Production-Optimized Congestion Control Selection for Alibaba Cloud CDN