C2DN: How to Harness Erasure Codes at the Edge for Efficient Content Delivery
Talkat 19th USENIX Symposium on Networked Systems Design and Implementation (NSDI '22) · 2022-04-04..04-06·
summary
Analyzes month-long logs from over 2000 clusters and implements C2DN on Apache Traffic Server, replacing replication with erasure coding plus targeted parity chunks to re-balance write load. Evaluation replaying production traces across three regions shows C2DN reduces byte miss ratio by 11%, eliminates unavailability-induced miss-ratio spikes, and cuts write-load imbalance by 99% while slightly increasing CPU/disk use; includes a max-flow parity-assignment solver and production-ready optimizations.
AI-written summary · about
chapters (18)
- 0:00Intro
- 0:27Content Delivery Networks CD
- 0:45Cache misses are expensive
- 1:46Unavailabilities at the edge are co
- 2:57Bucket-based routing coarser load ba
- 4:14State-of-the-art solution
- 5:05The problem: write load imbalan
- 7:59Quick primer on erasure coding: efficient fault
- 9:06Use erasure coding in CDN clusters
- 9:53Naive use of erasure coding is insuf chunks are evicted at different times
- 10:57Goal: rebalance the load
- 11:24Parity assignment problem
- 12:25Evaluation setup
- 12:45No more miss ratio spike
- 13:29Reducing normal case miss rat
- 14:02Near-perfect write load balancir
- 14:24In the paper
- 14:41Summary
related talks
01h 14mErasure Coding at Scale
12mCodedBulk: Inter-Datacenter Bulk Transfers using Network Coding
11mExploiting Combined Locality for Wide-Stripe Erasure Coding in Distributed Storage
38mUsing Erasure Codes to Enable Interactions over Networks
15mFrom Chaos to Clarity: Deciphering Cache Inconsistencies in a Distributed...
26mImplementing a Resilient Ceph RGW Object Storage with Erasure Coded Data Pool as the Foundation...
44mAvailability, Latency and Cost: Withstanding Outages
15mSelectiveEC: Selective Reconstruction in Erasure-coded Storage Systems