Kerveros: Efficient and Scalable Cloud Admission Control
summary
Kerveros is a Microsoft Azure admission-control system that uses late binding of reservations plus a novel bookkeeping primitive allocable VM to convert multi‑dimensional VM requirements into single counts. It runs fast allocation workers (zonal handlers, in‑memory placement snapshots, pub/sub) to sustain >120,000 requests/min with latencies of a few milliseconds, and a slow emulation + ML‑calibration loop (CRA and LAA run per minute and ~30 minutes) to reach >99.99% availability and much higher packing efficiency than partitioning or immediate allocation.
AI-written summary · about
chapters (9)
- 0:00Introduction to Kerberos and cloud challenges
- 2:17Admission control in Azure
- 5:03Late binding core idea
- 7:30Allocable VM accounting
- 9:54Conversion Ratio Algorithm
- 10:55Linear Adjustment Algorithm
- 12:19Alternatives comparison
- 13:43Resource efficiency results
- 14:25Production scalability
related talks
20mManage Scaling, While Keeping Costs Under Control. Challenge Accepted.
20mThe Coral Reef Cloud: Autoscaling, Reservation, and Preemption
30mCapacity Management for Fun & Profit
26mIntro to Cloud Provider Azure
48mBuilding applications at hyper scale with the latest Azure innovations
20mAzure: Maxmise your bang for Buck
27mCloudMirror: Application-Aware Bandwidth Reservations in the Cloud
53mFlex Consumption: Build Serverless Apps with High-Throughput and Virtual Networking Features