From I/O To RAM
Download as PPTX, PDF•6 likes•1,175 views
The document discusses various strategies and tools for managing memory and databases in application development, focusing on Terracotta's RAM patterns and caching solutions. It explores performance issues, scalability challenges, and presents several in-memory tools and frameworks such as memcached and Prevayler. Additionally, it details how to integrate Terracotta for efficient data handling, while highlighting its advantages and potential drawbacks.
From I/O To RAM
- 1. Riding The Memory BusOphir Radnitz
- 2. AgendaRAM vs. IOTools of the TradeIntroducing TerracottaRAM Patterns with Terracotta
- 3. Anatomy of an ApplicationGet a DB connection, start transactioningStart a requestNetworkGet a bunch of dataAccess disk,SerializationRelease resourcesRender stuff
- 4. I/O IssuesScalability issuesExcessive network chatterHow do you scale the database?Only as fast as your slowest queryImplies serialization/deserialization
- 5. Bandwidths (1 of 2)
- 6. Bandwidths (1 of 2)
- 7. RAM vs. DiskInternet: ~80ms
- 8. RAM GoodnessRAM = High Bandwidth & Low LatencyNo serialization costsLess network
- 9. Use CasesLow level cacheSQL, HTTPApplication cacheHibernate 2nd level cacheShort lived entitiesQueues
- 10. Generic ToolsMemcachedA high-performance, distributed memory caching systemUsed EVRYWHERE: Facebook, Twitter, Digg, Wikipedia, Slashdot, LiveJournal, Sourceforge etc.SquidRouting and load balancing HTTP cache
- 11. Memcached Limitations for JavaRemote cacheSerialization2Gb limit
- 12. Java In-Memory ToolsPrevayler, Space4JPersist objects in memory, supports ACID and file system journalingDb4O in memory, PerstObject oriented databaseJoftiA high-performance object indexing and searching solution. Supports Map.
- 13. In-Memory ConsiderationsReliability requires redundancyAnd/or durabilityIn a cluster state must be syncedRequires state distribution
- 15. Introducing TerracottaNetwork-Attached Memory infrastructure for the JVMOpen source(TPL, based on Mozilla License)Current version: 3.0.1
- 17. Introducing TerracottaHighly AvailableHighly ScalableAvoids excessive state replicationHub and spoke architecture
- 19. How Terracotta WorksApplicationTerracottaClientBootjarinstrumenting your applicationTerracotta LibrariesJVMTerracottaServerTC handles JVM locks and data access
- 20. How Terracotta Works, take 2ApplicationApplicationApplication Terracotta ServerReferenceActual instance
- 21. Terracotta FeaturesCross-JVM object identity Automatic Persistence to diskVirtual Memory spill heap to TC and/or to disk, transparentlyCluster profiling & visualizationJMX-based monitoring and management
- 22. Terracotta AdvantagesEasy to integrate, Maven supportGreat scalabilityNo proprietary API (optional)Ready made integration modulesGreat diagnosticsIn/out of process
- 24. Integrating TerracottaDownload & extractWrite a tc-config.xmlCreate a bootjarRun your app with the bootjarAdd elements & refactor
- 25. TIM - Terracotta Integration ModulesPackaged functionalityEasy to integrateHandles locks detailsExamples:Concurrent collections, EhcacheLucene/Compass, Spring SecurityPipes, Wicket, Tomcat, Jetty Hibernate, JBoss, Spring etc.
- 26. Terracotta ElementsHTTP Session ReplicationCache EvictorQueuesMaster-WorkerAsynchronous ProcessorCluster membership events
- 27. Terracotta PatternsData CacheIntermediate persistenceQueues / work distributionWrite behind to database
- 28. Data Cache ConsiderationsFigure out data lifecycleRead-only, metadata, configuration, sessionWhere cache is applicableSingle point of updateSet eviction policiesLRU, FIFO
- 29. Data CacheSession cache is easier, entity cache needs real distributionUse a generic cacheConcurrentMap + Map Evictor (+ Jofti)Or a Hibernate 2nd level cacheehcache, TC-Cache
- 30. Where Should My Objects Live?MemoryDatabaseAppropriatenessdies quicklystays foreverData Lifetime
- 31. Intermediate Data LifetimeSome data doesn’t have to be stored in the databaseSession informationMessages / TokensTransactionUnfinished stuffData Lifetime
- 32. TIM Pipes: QueuesExecute time consuming tasks asynchronouslyIn a reliable wayLoad balancedExamples:Send emails, process images Supports pub/sub, routingNo need for JMS
- 33. CommonJ – Work DistributionA joint API spec by Oracle and IBMProvides interfaces for:Worker / WorkItem / WorkManagerScheduleretc…TIM-Messaging provides an implementationPipesMaster / Worker
- 34. Write Behind to System Of RecordAsynchronously persist to DBWork queue“1.5 phase commit”Flags objects as dirtyAvoids syncing dirty objectsFlushes objectsFlags objects as clean
- 35. Terracotta DrawbacksConcurrency awarenessMassive scalability is a commercial featureRequires another machine(s)Sharing objects across applications requires configurationTIMs documentation
- 36. Terracotta ProsConcurrency awarenessFits into your applicationProvides great visibilityVery versatile usageScales massivelyDoesn’t use serializationGreat community
- 37. SummaryAsk yourself:Does it belong in the DB?Do I have to go to the DB for that?Do I have to do it now?