1.JuiceFS Juicedata 2018.11.18 

2. DPark Apache Spark MooseFS HDFS JuiceFS Beansdb 2008 2018 Douban Facebook Databricks Juicedata 

3.• Cloud Native • • 

4.• • S3, OSS • HDFS 

5. (S3) • API • Restful • • 

6. (S3) • • • FileNotFoundException • • • : FileAlreadyExistsException • 

7. (EFS) S3 EFS 300 225 150 75 0 (/TB/ 

8. HDFS • • • 

9. HDFS • • • CPU • • 

10.• • • • • • 

11. HDFS ZooKeeper DataNode1 JN1 DataNode2 JN2 NameNode JN3 DataNode3 NN Standby HDFS Client (JVM) 

12. HDFS ZooKeeper Data DataNode1 Manager JN1 DataNode2 JN2 Meta server JN3 DataNode3 NN Standby HDFS Client (JVM) 

13. HDFS ZooKeeper DN DataNode1 Manager JN1 DataNode2 JN2 Meta server JN3 DataNode3 NN Standby HDFS Client (JVM) 

14. JuiceFS S3 GCS OSS Ceph Meta Service meta data Juicedata kernel/FUSE 

15. File chunk 64MB 64 bits id object 4MB /chunks/1/1024/1024123_0_4194304 /foo /chunks/1/1024/1024123_1_4194304 /chunks/1/1024/1024123_2_256 

16. Raft Group Follower Follower Logger changelog Logger Leader Logger Logger SSL 

17. JuiceFS • • Raft • • key 

18. JuiceFS • • • • • • • • 

19. JuiceFS 300 225 150 75 0 S3 EFS HDFS JuiceFS 

20. JuiceFS • • • • 

21.S3 OSS 

22. <1s < 30s S3 OSS 

23. JuiceFS • • • • • • 

24. Flink Spark TensorFlow Hive JuiceFS MPI HBase Impala 

25. Thanks Q&A davies@juicedata.io