⾼高性能分布式PaaS解决⽅方案 ⺩王璞@数⼈人科技 2015/10/17 Why • Why high performance? • • Business driven • More users, more data, etc • Scalability Why PaaS? • Facilitate building cloud native application • Better leverage cloud computing to achieve high performance Unreliable Hardwares • Cost Platform-as-a-Service Spark Docker Application Jenkins Kafka Cassandra Log Management Load Balancer Mesos Service Discovery Monitoring Cloud Native Application • Fault tolerant • Due to unreliable hardwares • No single point failure • Auto recover • • Stateless, easy to recover Extensibility • Due to fast business growth • Keep application architecture unchanged while adding recourses • Micro-service, easy to extend How Cloud Native Requirements Application + PaaS Fault Tolerant Mesos + Marathon Stateless Application + DB/MQ/… Extensibility Mesos + Marathon Micro-service Application + Docker Mesos • Multiple masters, synchronized by Zookeeper • ~1,000 Slaves, each 2 core 4G • Frameworks: Marathon, Chronos, Spark, Hadoop, Storm, Kafka, Cassandra, ElasticSearch Job Running - Shared Resource Job Scheduling - Dynamic Scheduling Job Scheduling - Dynamic Scheduling Job Scheduling - Fault Tolerance Docker • Lightweight • Portability • Immutable infrastructure • Easy updates VM v.s. Docker VM v.s. Docker Extensibility • Keep application architecture unchanged while adding recourses Service Extensibility - Load Balancing Service Extensibility - Service Discovery Data Extensibility - Sharding