Scaling Up Hibernate/JPA Applications with Infinispan Second-Level Cache

Transcript

2. Scaling Up Hibernate/JPA Applications with Infinispan Second-Level Cache ! Galder Zamarreño Senior Software Engineer Red Hat, Inc ! 20th April 2013, São Paulo
3. Galder Zamarreño • R&D Engineer, Red Hat Inc. • Infinispan developer • Creator of Infinispan second-level cache provider for Hibernate • Escalante founder and lead • Twitter: @galderz
4. Agenda • Bottlenecks in Hibernate/JPA applications • Applying caching techniques sensibly • Distribute caches across multiple nodes • Caching in managed environments
5. Why do we care?
6. DB = bottleneck Most Hibernate/JPA applications involve intensive I/O workload against database
7. Bottleneck problems Bottlenecks decrease throughput, increase latency... and make users :(((
8. We care because we want to make users :))
9. How do we scale up Hibernate/JPA applications ?
10. Option 1: $$$ Buy/rent more powerful hardware to host the DB... who can afford that?
11. Option 2: Caching Use second-level caching sensibly to reduce burden on database
12. Second Level Cache JVM-level, or clustered, cache whose contents can be shared between different transactions
13. Lesson 1: Don't apply caching blindly! Always measure first!
14. Caching cost! Local caches have a cost (memory, GC), clustered caches add more cost (network, IO)
15. Demo 1: Measure performance
16. Lesson 2: Think what you are going to cache!
17. Think about reads Biggest performance gain comes from caching read- mostly data
18. What kind of data? Entities, collections of entities, and query results can be cached
19. Entity caching Mark entities cacheable, and define the cache concurrency strategy
20. Query caching Speed up execution of repeated queries (w/ same parameters) by caching their results
21. Query results validity Cached query results are valid as long as none of the entity types involved are updated
22. Demo 2: Query cache in action
23. Lesson 3: How to select a caching provider
24. Cache Strategies • Read-only • data never updated • insert/delete allowed • Read/write • data can be updated • with JBDC transactions or no transactions
25. Cache Strategies • Nonstrict read/write • data rarely updated • with JBDC transactions or no transactions • Transactional • data can be updated • for JTA environments
26. Strategy Support Provider / Strategy read- only nonstrict r/w r/w transactional Infinispan yes no no yes EhCache yes yes yes yes ConcurretHashMap (testing) yes yes yes no
27. Lesson 4: Scale up your Hibernate/JPA app going multi-node!
28. Caches form cluster Infinispan uses JGroups to form a peer-to-peer cluster, providing discovery, failure detection...etc
29. Clustered Entities Recommended using invalidation to keep entities consistent accross cluster (replication can be used too)
30. Queries / Timestamps Recommended keeping query cache local to each node, but must replicate update timestamps cache
31. Demo 3: Clustered second-level cache in action!
32. Lesson 5: Scaling up JPA in managed environments
33. JPA Caching simplified Infinispan as default cache provider (with sensible defaults), and ready to cluster!
34. shared-cache-mode Persistence unit configuration option that should be set to ENABLE_SELECTIVE
35. Demo 4: JPA caching in JBoss AS 7
36. Summary Q - How do we scale up Hibernate/JPA applications? Enable second level cache!!
37. Summary • Don't apply caching blindly! Always measure first! • Think what you are going to cache! • How to select a caching provider • Scale up your app going multi-node! • Scaling up JPA in managed environments
38. Questions • http://infinispan.org • http://hibernate.org • http://github.com/galderz/secondlc • https://docs.jboss.org/author/display/ ISPN/Using+Infinispan+as+JPA-Hibernate +Second+Level+Cache+Provider