Virtualization at Scale @ Alibaba Cloud
summary
Alibaba describes Apsara as a hyperscale cloud OS powering 5,000-node clusters (now exceeding 10,000 servers with hundreds of PB and hundreds of thousands of vCPUs) and reports 87 million paying customers and ¥6.6B revenue in FY2017. Engineering highlights: a world‑record 100 TB sort in 377 seconds, transition from Xen to KVM with hot KVM upgrades with millisecond pauses, CPU isolation that raises utilization from ~35% to >65% with <5% latency loss, network/io throttling (250k IOPS threshold) and support for Docker Swarm and Kubernetes.
AI-written summary · about
chapters (13)
- 0:00Intro
- 1:20Apsara: A brief History
- 4:53Planet-scale Infrastructure
- 6:15Apsara: A Hyper-scale Cloud Operating System
- 9:57Apsara Design Highlights
- 11:56Apsara 5K
- 15:28Virtualization in General
- 16:41Mixing Latency-sensitive (LS) and Batch Workload
- 19:40Server Virtualization
- 20:16Hypervisor Hot Upgrade
- 21:34Alibaba Cloud Loves Containers
- 22:10Scalable Container Service on Cloud
- 23:09Ideas of Future Directions
related talks
22mKeynote: Cloud Native at Alibaba Scale
41mCloud Computing with KVM
22mKeynote: Cloud Native at Alibaba Scale
16mThe Supercloud
01h 13mEngineering the Cloud for Tomorrow: Performance, Speed and Scale for AI Workloads
18mProgrammable Hyperscale Cloud
31mAccelerate AI adoption: automated cloud efficiency
48mBuilding applications at hyper scale with the latest Azure innovations