Advances, Systems and Applications
From: Online architecture for predicting live video transcoding resources
Parameter | Description |
---|---|
Video source | Online live video from Twitch. Resolution: 1280 × 720, Video format: HLS (segment length: 1 s) |
Prediction algorithms | RF, RL, RL with multiple RL agents (4) |
Reference | vCPU request = x%*vCPU limit |
Online model training | Every 30 min |
Target | Transcoding speed > 1.0 |
vCPU limit | 2 vCPU = 256 × 144 (197 kbps), 3 vCPU = 640 × 360 (829 kbps), 4 vCPU = 854 × 480 (1380 kbps) |
vCPU request | 35%*vCPU limit for VM with 4 vCPU, 45%*vCPU limit for VM with 6 and 8 vCPU, 50%*vCPU limit for VM with 16 vCPU |
Video URL source | Twitch Stream API (v5) [55] |
Video URL conversion | Streamlink [56] |