Big data software - IMWorld 2012

Real time analytics is a compelling use case for NoSQL solutions like MongoDB, Cassandra and CouchDB, but only if everything is up and running smoothly. We'll look at and compare each of these data stores alongside case studies for each. This will cover what to consider for high write throughput performance from hardware configuration through to the use of replication, multi-data centre deployments, monitoring and sharding to ensure your database is fast and stays online.

Presented at http://www.imworld.ro 2012 by David Mytton

Transcript

1. Big data software
2. David Mytton Woop Japan!
6. Similarities
7. Similarities • Document stores
8. Similarities • Document stores • Flexible schema
9. Similarities • Document stores • Flexible schema • Purpose
10. Similarities • Document stores • Flexible schema • NoSQL • Purpose
11. It’s a little different.
12. Differences
13. Differences • Implementation
14. Differences • Implementation • Durability
15. Differences • Implementation • Durability • Queries
17. Scaling writes
18. • Global lock Scaling writes
19. • Global lock • Concurrency Scaling writes
20. http://bit.ly/mongolock
21. • Global lock • Sharding • Concurrency Scaling writes
22. Scaling writes
23. Scaling reads
24. •Replica slaves Scaling reads •Consistency •w flag / tags
25. http://www.flickr.com/photos/comedynose/4388430444/ Bottlenecks www.flickr.com/photos/comedynose/4388430444/
26. http://www.flickr.com/photos/comedynose/4388430444/ Bottlenecks •RAM www.flickr.com/photos/comedynose/4388430444/
27. http://www.flickr.com/photos/comedynose/4388430444/ Bottlenecks •RAM www.flickr.com/photos/comedynose/4388430444/ What? Should it be in memory? Indexes Always Data If you can
28. Bottlenecks •Disk i/o www.flickr.com/photos/daddo83/3406962115/
29. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks •Disk i/o •Mount points
30. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks •Disk i/o •SSD •Mount points
31. Failover •Replica sets
32. Failover •Replica sets •Master/slave
33. Failover •Replica sets •Min 3 nodes •Master/slave
34. Failover •Replica sets •Min 3 nodes •Master/slave •Automatic failover
35. Redundancy •Tags { _id : "someSet", members : [ {_id : 0, host : "A", tags : {"dc": "ny"}}, {_id : 1, host : "B", tags : {"dc": "ny"}}, {_id : 2, host : "C", tags : {"dc": "sf"}}, {_id : 3, host : "D", tags : {"dc": "sf"}}, {_id : 4, host : "E", tags : {"dc": "cloud"}} ] settings : { getLastErrorModes : { veryImportant : {"dc" : 3}, sortOfImportant : {"dc" : 2} } } } > db.foo.insert({x:1}) > db.runCommand({getLastError : 1, w : "veryImportant"})
36. Case Study
37. Case Study •Server Density •26 nodes •6 replica sets •Primary datastore = 15 nodes
38. Case Study •Server Density •+12TB / mth •+1bn docs / mth •3-5k inserts/s @ 0.1ms
40. Scaling writes
41. Scaling writes 1) Commit log
42. Scaling writes 1) Commit log 2) Memtable
43. Scaling writes 1) Commit log 2) Memtable 3) Disk - sequentially
44. Scaling writes Image: www.datastax.com
45. Scaling writes Image: www.datastax.com
46. Scaling writes Image: www.acunu.com
47. Scaling writes Image: www.acunu.com
48. Scaling reads
49. Scaling reads •Many SSTables
50. Scaling reads •Many SSTables •Locate the right one(s)
51. Scaling reads •Many SSTables •Locate the right one(s) •Fragmentation
52. Scaling reads Image: www.acunu.com
53. http://www.flickr.com/photos/comedynose/4388430444/ Bottlenecks •RAM www.flickr.com/photos/comedynose/4388430444/
54. http://www.flickr.com/photos/comedynose/4388430444/ Bottlenecks •RAM www.flickr.com/photos/comedynose/4388430444/ •Compression • 2x-4x reduction in data size • 25-35% performance improvement on reads • 5-10% performance improvement on writes
55. Bottlenecks •Node size
56. Bottlenecks •Node size •Startup time
57. Failover •Replication
58. Failover •SimpleStrategy Image: www.datastax.com
59. Failover •NetworkTopologyStrategy Image: www.datastax.com
60. Case Study
61. Case Study •Britain’s Got Talent •2 nodes •10k votes/s •RDS m1.large = 300/s
63. www.ex-astris-scientia.org/inconsistencies/ent_vs_tng.htm (yes it’s a replicator from Star Trek) Scaling
64. www.ex-astris-scientia.org/inconsistencies/ent_vs_tng.htm (yes it’s a replicator from Star Trek) •Replication Scaling
65. www.ex-astris-scientia.org/inconsistencies/ent_vs_tng.htm (yes it’s a replicator from Star Trek) •Replication Scaling •Replication
66. www.ex-astris-scientia.org/inconsistencies/ent_vs_tng.htm (yes it’s a replicator from Star Trek) •Replication Scaling •Replication •Replication
67. Picture is unrelated! Mmm, ice cream. Scaling
68. Picture is unrelated! Mmm, ice cream. •HTTP Scaling
69. Picture is unrelated! Mmm, ice cream. •HTTP Scaling •Load balancer
70. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks
71. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks •Disk space
72. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks •Disk space •No ad-hoc
73. www.flickr.com/photos/daddo83/3406962115/ Bottlenecks •Disk space •No ad-hoc •Append only
74. Failover
75. Failover •Replication
76. Failover •Replication •Eventual consistency
77. Failover •Replication •Eventual consistency •DNS
78. DIY
79. DIY •Replication
80. DIY •Replication •Failover
81. DIY •Replication •Failover •Queries
82. Case Study
83. Case Study •BBC •8 nodes per DC •Eventual consistency •DNS failover •8 nodes per DC
84. Case Study •BBC •Max 1k PUT/s/node •24 PUT/s •500 GET/s
85. Case Study •BBC •No direct access •Caching •k/v store
86. Credits •Tom Wilkie, Acunu •Simon Lucy, BBC
87. David Mytton david@serverdensity.com @davidmytton Woop Japan! www.serverdensity.com