SHLUG 10 月月度聚會速記 - 高性能分佈式 PaaS 解決方案
- Why high performance
- bisuness driivr
- messos
- developered in c++
- maintained by apache
- log mangement
- monitoring
- load balancer
- service discovery: 自動組態一群 services, 無人參與
- cloud native application
- fault tolerant
- due to unreliable harewrres
- no single point failur
- auto recover
- staeless, easy to recover
- extensibility
- 處理突發大量需求
- due to fast buniess grouth
- keep application architecture unchaged while adding resources
- micro-service, easy to extend
- fault tolerant
- How
cloud native reuirements | application + PaaS
fault tolernt | msos + marathon
stateless | application db/mq/...
extensibility | mesos + marathon
micro-service | application + docker/kvm/xen - hyper
- mesos
- multiple masters, synchronized by zookeeper
- 1000+ slaves, each 2 core 4G
- frameworks: marathon, chronos, spark, hadoop, storm, kafka, sa...
- bork?
- master have HA supporting, one died, select by election
- application can scale up by invoke framework and release resource by itself
- 目前資源配置沒有 priority 機制
- http://www.shurenyun.com