Downloaded 593 times
![Inserting: OverviewSimple: put(key, col, value) Complex: put(key, [col:value, …, col:value]) Batch: multi key.](https://image.slidesharecdn.com/cassandrasvccg10june2010-100618145003-phpapp02/85/Introduction-to-Cassandra-June-2010-36-320.jpg)
![Inserting: OverviewSimple: put(key, col, value) Complex: put(key, [col:value, …, col:value]) Batch: multi key.](https://image.slidesharecdn.com/cassandrasvccg10june2010-100618145003-phpapp02/75/Introduction-to-Cassandra-June-2010-36-2048.jpg)
More Related Content
![[PASS Summit 2016] Azure DocumentDB: A Deep Dive into Advanced Features](https://cdn.slidesharecdn.com/ss_thumbnails/passdocdbadvfeatures-161029023827-thumbnail.jpg?width=600ounds&width=560&fit=bounds)
Similar to Introduction to Cassandra (June 2010)
Introduction to Cassandra (June 2010)
- 1. ApacheGary DusbabekRackspaceSilicon ValleyCloud Computing Group • 17 June 2010
- 2. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 3. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 4. Why Cassandra?1.98 billion500 GB drives988EB6 fold growthIn 4 years322 million 500GB drives161 EB20062010Source: http://www.emc.com/collateral/analyst-reports/expanding-digital-idc-white-paper.pdf
- 5.
- 6. SQLSpecialized data structures(think B-trees)Shines with complicated queriesFocus on fast query & analysis quicklyNot necessarily on large datasets
- 7. Ever tried scalinga RDBMSFor reads?Memcache etc.For writes?Oh noes!
- 8. VerticalScalingIs hardcredit: janetmckvia flickr
- 9.
- 10. Enter CassandraAmazon DynamoConsistenthashingPartitioningReplicationOne-hop routingGoogle BigTableColumn FamiliesMemtablesSSTables
- 11.
- 12.
- 13.
- 14. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 15. Distributed and ScalableHorizontal!Allnodes are identicalNo master or SPOFAdding is simpleAutomatic cluster maintenance
- 16. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 17. ReplicationReplication factorHow manynodes data is replicated onConsistency levelZero, One, Quorum, AllSync or async for writesReliability of readsRead repair
- 18. Ring TopologyRF=3Conceptual RingOnetoken per nodeMultiple ranges per nodeajdg
- 19. Ring TopologyRF=2Conceptual RingOnetoken per nodeMultiple ranges per nodeajdg
- 20. New NodeRF=3Token assignmentRangeadjustmentBootstrapArrival only affects immediate neighborsamjdg
- 21.
- 22. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 23. Schema-free Sparse-tableFlexible columnnamingYou define the sort orderNot required to have a specific column just because another row does
- 24.
- 25.
- 26.
- 27. Easier to showfrom the bottom up
- 28. Data ModelA singlecolumn
- 29.
- 30.
- 31. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 32. Eventually ConsistentCAP TheoremConsistencyAvailabilityPartitionToleranceChoose twoCassandra chooses A and PBut…
- 33. Eventually ConsistentI gota fever! And the only prescription is MORE CONSISTENCY!
- 34. Tunable ConsistencyGive upa little A and P to get more CRatchet up the consistency levelR + W > N Strong consistencyMore to come
- 35. OutlineHistoryScalingReplication ModelData ModelTuningWritePathRead PathClient AccessPractical Considerations
- 36. Inserting: OverviewSimple: put(key,col, value) Complex: put(key, [col:value, …, col:value]) Batch: multi key.
- 37. Inserting: WritesCommit logfor durabilityConfigurable fsyncSequential writes onlyMemtable – no disk access (no reads or seeks)Sstables are final (become read only)IndexesBloom filterRaw dataBottom line: FAST!!!
- 38. OutlineHistoryScalingReplication ModelData ModelTuningWritePathReadPathClient AccessPractical Considerations
- 39. Querying: OverviewYou needa key or keys:Single: key=‘a’Range: key=‘a’ through ’f’And columns to retrieve:Slice: cols={bar through kite}By name: key=‘b’ cols={bar, cat, llama}Nothing like SQL “WHERE col=‘faz’”But secondary indices are being worked on (see CASSANDRA-749)
- 40. Querying: ReadsPractically lockfreeSstable proliferationNew in 0.6:Row cache (avoid sstable lookup, not write-through)Key cache (avoid index scan)
- 41. OutlineHistoryScalingReplication ModelData ModelTuningWritePathReadPathClient AccessPractical Considerations
- 42. Client API (LowLevel)Fat ClientLive non-storage nodeReduced RPC overheadThrift (12 language bindings!)http://incubator.apache.org/thrift/No streamingAvroWork in progress
- 43. Client API (HighLevel)http://wiki.apache.org/cassandra/ClientOptions
- 44. Feature richConnection poolingLoadbalancing/failoverSimplified APIsVersion opaque
- 45. OutlineHistoryScalingReplication ModelData ModelTuningWritePathReadPathClient AccessPractical Considerations
- 46. Practical Considerations Partitioner-Random orOrder PreservingRange queriesProvisioningVirtual or bare metal Cluster sizeData modelThink in terms of accessGiving up transactions, ad-hoc queries, arbitrary indexes and joins(you may already do this with an RDBMS!)
- 47. Practical ConsiderationsWide rowsDatalife-spanCluster planningBootstrapping
- 48. Future DirectionVector clocks(server side conflict resolution)Alter keyspace/column families on a live clusterCompressionMulti-tenant featuresLess memory restrictions
- 49. Wrapping UpUse Cassandraif you want/needHigh write throughputNear-linear scalabilityAutomated replication/fault toleranceCan tolerate missing RDBMS features
- 50. Questions?Linkagewiki.apache.org/cassandracassandra.apache.orggdusbabek@gmail.comgdusbabek on twitterand just about everything else.
Editor's Notes
- #5 Data growth has been expanding.
- #6 Historical industry leaders
- #9 32 core processor machines are expensiveCosts go way up when you try to scale these databasesAlso-instability.
- #10 Terabytes of data~1,000,000 ops/secondSchema changes are difficult (impossible)Manual sharding takes a lot of effortAutomated sharding + replication is difficult
- #12 100 M users, 25 TB data
- #16 Horizontal – commodity hardware, not specialized boxes
- #19 Cluster is a logical storage ringNode placement divides the ring into ranges that represent start/stop points for keysAutomatic or manual token assignment (use another slide for that) Closer together means less responsibility and data
- #20 Token
- #21 Bootstrapping
- #22 Hinting not designed for long failures.
- #32 RDBMS focus on consistency. Limits scale.
- #35 No multi-key transactions
- #39 Sstable proliferation degrades performance.
- #47 DistributedScalableSchema-freeSparse tableEventually consistentTunable (throughput and fault-tolerance)