Database High Availability Using SHADOW Systems
1 like210 views
The document presents SHADOW, a system for providing high availability of databases using shared persistent storage in cloud environments. SHADOW maintains a single logical copy of the database and log across active and standby database management systems. It pushes the responsibility for data replication into the underlying storage tier. This decouples database replication from replication at the DBMS level. The document evaluates SHADOW against standalone and synchronous replication baselines using the TPC-C benchmark. SHADOW outperforms the other approaches and provides more stable throughput over time by offloading write operations and exploiting data sharing across nodes. However, its replication is limited by the scope of the shared storage.
Database High Availability Using SHADOW Systems
- 1. Database High Availability Using SHADOW Systems Jaemyung Kim, Kenneth Salem, Khuzaima Daudjee, Ashraf Aboulnaga, and Xin Pan University of Waterloo SoCC 2015
- 2. SHADOW Hot Standby HA for Cloud How can we exploit shared persistent storage to build better highly available database systems? 2
- 3. Overview 1 Standalone and Hot Standby Failure Handling 2 Shared Storage in Cloud Settings 3 SHADOW: Hot Standby HA for Cloud 4 Performance Evaluation 5 Conclusion 3
- 4. Example DBMS Setting x DBMS bu↵er pool Log DB C 4
- 5. Standalone Restart Recovery x DBMS bu↵er pool Log DB C node failure 4
- 6. Standalone Restart Recovery x DBMS bu↵er pool Log DB C restart recovery 4
- 7. Typical Shared-Nothing Hot Standby Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB x C C 5
- 8. Hot Standby Failure and Failover Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB x node failure C C 5
- 9. Hot Standby Failure and Failover Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB x node failure failover C C 5
- 10. Hot Standby Is Widely Used Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB x node failure C C 5
- 11. Typical Hot Standby High Availability Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB 6
- 12. Shared Storage in Cloud Settings Storage in Cloud? Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB 6
- 13. Shared Storage in Cloud Settings Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB Persistent Storage = Reliable Shared Storage 6
- 14. Shared Storage in Cloud Settings Active DBMS bu↵er pool Log DB Standby DBMS bu↵er pool Log DB How can we exploit shared persistent storage to build better highly available database systems? 6
- 15. SHADOW: Hot Standby HA for Cloud Active DBMS bu↵er pool Standby DBMS bu↵er pool DB DBLog Log Recycled Hot Standby HA in Cloud 7
- 16. SHADOW: Single Logical Log Active DBMS bu↵er pool Standby DBMS bu↵er pool DB DBLogLog async rep Single Logical Log 7
- 17. SHADOW: Single Logical Database Active DBMS bu↵er pool Standby DBMS bu↵er pool DB DBLogLog Xwrite-o✏oading coordinated checkpoint Single Logical Database 7
- 18. SHADOW: Hot Standby HA for Cloud Active DBMS bu↵er pool Standby DBMS bu↵er pool x DBLog SHADOW High Availability 7
- 19. SHADOW: Hot Standby HA for Cloud Active DBMS bu↵er pool Standby DBMS bu↵er pool x DBLog C 7
- 20. SHADOW: Node Failure and Failover Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C 7
- 21. SHADOW: Node Failure and Failover Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure failover DBLog C 7
- 22. Advantages of SHADOW HA Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C Simplicity: pushes responsibility for durability and replication into the storage system 7
- 23. Advantages of SHADOW HA Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C Simplicity: pushes responsibility for durability and replication into the storage system Flexibility: decouples database replication from DBMS replication 7
- 24. Advantages of SHADOW HA Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C Simplicity: pushes responsibility for durability and replication into the storage system Flexibility: decouples database replication from DBMS replication Performance: write-o✏oading (logging and checkpointing) 7
- 25. Advantages of SHADOW HA Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C Simplicity: pushes responsibility for durability and replication into the storage system Flexibility: decouples database replication from DBMS replication Performance: write-o✏oading (logging and checkpointing) E ciency: less I/O bandwidth 7
- 26. Advantages of SHADOW HA Active DBMS bu↵er pool Standby DBMS bu↵er pool x node failure DBLog C Simplicity: pushes responsibility for durability and replication into the storage system Flexibility: decouples database replication from DBMS replication Performance: write-o✏oading (logging and checkpointing) E ciency: less I/O bandwidth 7
- 27. Experimental Methodology Compare SHADOW System with two baselines Standalone (SA): single DBMS with restart recovery (varying checkpint interval) Synchronous Replication (SR): two replicated DBMSes (native PostgreSQL implementation) TPC-C Benchmark (100 Warehouses), no think time PostgreSQL 9.3 (database fits in memory) Linux kernel 3.2.0-56 Amazon EC2 with Elastic Block Store (EBS) 8
- 28. TPC-C Benchmark Throughtput 0 10000 20000 30000 40000 50000 60000 SAD SA10 SA Throughput(tpmC) Standalone Amazon EC2 c3.4xlarge instances, 100WH TPC-C workload database fits in memory (PostgreSQL 9.3, Linux kernel 3.2.0-56) 9
- 29. TPC-C Benchmark Throughtput 0 10000 20000 30000 40000 50000 60000 SAD SA10 SA SR Throughput(tpmC) Standalone Amazon EC2 c3.4xlarge instances, 100WH TPC-C workload database fits in memory (PostgreSQL 9.3, Linux kernel 3.2.0-56) Hot Standby 9
- 30. TPC-C Benchmark Throughtput 0 10000 20000 30000 40000 50000 60000 SAD SA10 SA SR SHADOW Throughput(tpmC) Standalone Amazon EC2 c3.4xlarge instances, 100WH TPC-C workload database fits in memory (PostgreSQL 9.3, Linux kernel 3.2.0-56) Hot Standby 9
- 31. Variability of TPC-C Throughput 0 10000 20000 30000 40000 50000 60000 SAD SA10 SA tpmC Standalone Box-and-whisker plot (Q1,Q2,Q3) Ten second interval (new order transactions per second x 60)) 0 10000 20000 30000 40000 50000 60000 SAD SA10 SA SR SHADOW tpmC Hot Standby 10
- 32. Conclusion SHADOW Hot Standby HA for Cloud How to exploit shared storage to build better high available database systems? Single logical copy of the database and log Pushes responsibility for replication out of the DBMS and into the underlying storage tier Decouples database replication from DBMS replication Outperforms PostgreSQL’s native replication on a TPC-C benchmark Stabler throughput(tpmC) over time Geographical limitation Replication coverage is limited to shared storage coverage 11
- 33. Mahalo! 12