Midgress-aware traffic provisioning for content delivery
summary
Proposes midgress-aware traffic provisioning that assigns hundreds of traffic classes to CDN servers to cut cache-miss (midgress) traffic; uses footprint descriptors and FFT-based convolution to predict miss-rate curves and eviction ages, formulates an MILP for optimal assignment, and implements a fast local-search heuristic that runs quickly at production scale. Evaluation on 25 real traffic classes from Akamai shows near 20% midgress reduction versus midgress-unaware baselines, with the heuristic close to optimal and practical for deployment.
AI-written summary · about
chapters (11)
- 0:00Start
- 1:29Performance and cost metrics
- 3:04Cache management
- 4:52Traffic provisioning to reduce midgress
- 6:22Eviction age equality
- 7:16Footprint descriptors
- 8:47Traffic mixing using FD calculus
- 9:21Traffic provisioning to minimize midgress
- 9:59Traffic provisioning as an optimization problem
- 12:13Metro-level traffic provisioning Traffic classes
- 13:05Trace characteristics
related talks
16mMedley: Optimizing Midgress Bandwidth for Commercial Live Streaming CDNs
19mLearning Production-Optimized Congestion Control Selection for Alibaba Cloud CDN
16mCredence: Augmenting Datacenter Switch Buffer Sharing with ML Predictions
12mCost-effective Cloud Edge Traffic Engineering with Cascara
17mTowards provably performant congestion control
16mOptimizing Network Provisioning through Cooperation
20mPostMan: Rapidly Mitigating Bursty Traffic by Offloading Packet Processing
17mBridging the Gap between QoE and QoS in Congestion Control: A Large-scale Mobile...
