SlideShare a Scribd company logo

Big Data in Real-Time at Twitter

•Download as KEY, PDF•
•
N
nkallen

The document summarizes how Twitter handles and analyzes large amounts of real-time data, including tweets, timelines, social graphs, and search indices. It describes Twitter's original implementations using relational databases and the problems they encountered due to scale. It then discusses their current solutions, which involve partitioning the data across multiple servers, replicating and indexing the partitions, and pre-computing derived data when possible to enable low-latency queries. The principles discussed include exploiting locality, keeping working data in memory, and distributing computation across partitions to improve scalability and throughput.

1 of 71
Downloaded 7,663 times
Big Data in Real-Time at Twitter by Nick Kallen (@nk)
Follow along http://www.slideshare.net/nkallen/qcon
What is Real-Time Data? • On-line queries for a single web request • Off-line computations with very low latency • Latency and throughput are equally important • Not talking about Hadoop and other high-latency, Big Data tools
The four data problems • Tweets • Timelines • Social graphs • Search indices
Big Data in Real-Time at Twitter
What is a Tweet? • 140 character message, plus some metadata • Query patterns: • by id • by author • (also @replies, but not discussed here) • Row Storage
Find by primary key: 4376167936
Find all by user_id: 749863
Original Implementation id user_id text created_at 20 12 just setting up my twttr 2006-03-21 20:50:14 29 12 inviting coworkers 2006-03-21 21:02:56 34 16 Oh shit, I just twittered a little. 2006-03-21 21:08:09 • Relational • Single table, vertically scaled • Master-Slave replication and Memcached for read throughput.
Original Implementation Master-Slave Replication Memcached for reads
Problems w/ solution • Disk space: did not want to support disk arrays larger than 800GB • At 2,954,291,678 tweets, disk was over 90% utilized.
PARTITION
Possible implementations Partition by primary key Partition 1 Partition 2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... 24 ... 25 ...
Possible implementations Partition by primary key Partition 1 Partition 2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... 24 ... 25 ... Finding recent tweets by user_id queries N partitions
Possible implementations Partition by user id Partition 1 Partition 2 id user_id id user_id ... 1 21 2 ... 1 23 2 ... 3 25 2
Possible implementations Partition by user id Partition 1 Partition 2 id user_id id user_id ... 1 21 2 ... 1 23 2 ... 3 25 2 Finding a tweet by id queries N partitions
Current Implementation Partition by time id user_id 24 ... Partition 2 23 ... id user_id Partition 1 22 ... 21 ...
Current Implementation Queries try each Partition by time partition in order id until enough data user_id 24 ... is accumulated Partition 2 23 ... id user_id Partition 1 22 ... 21 ...
LOCALITY
Low Latency PK Lookup Memcached 1ms MySQL <10ms* * Depends on the number of partitions searched
Principles • Partition and index • Exploit locality (in this case, temporal locality) • New tweets are requested most frequently, so usually only 1 partition is checked
Problems w/ solution • Write throughput • Have encountered deadlocks in MySQL at crazy tweet velocity • Creating a new temporal shard is a manual process and takes too long; it involves setting up a parallel replication hierarchy. Our DBA hates us
Future solution Partition k1 Partition k2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... Partition u1 Partition u2 user_id ids user_id ids 12 20, 21, ... 13 48, 27, ... 14 25, 32, ... 15 23, 51, ... • Cassandra (non-relational) • Primary Key partitioning • Manual secondary index on user_id • Memcached for 90+% of reads
The four data problems • Tweets • Timelines • Social graphs • Search indices
Big Data in Real-Time at Twitter
What is a Timeline? • Sequence of tweet ids • Query pattern: get by user_id • Operations: • append • merge • truncate • High-velocity bounded vector • Space-based (in-place mutation)
Tweets from 3 different people
Original Implementation SELECT * FROM tweets WHERE user_id IN (SELECT source_id FROM followers WHERE destination_id = ?) ORDER BY created_at DESC LIMIT 20
Original Implementation SELECT * FROM tweets WHERE user_id IN (SELECT source_id FROM followers WHERE destination_id = ?) ORDER BY created_at DESC LIMIT 20 Crazy slow if you have lots of friends or indices can’t be kept in RAM
OFF-LINE VS. ONLINE COMPUTATION
Current Implementation • Sequences stored in Memcached • Fanout off-line, but has a low latency SLA • Truncate at random intervals to ensure bounded length • On cache miss, merge user timelines
Throughput Statistics date average tps peak tps fanout ratio deliveries 10/7/2008 30 120 175:1 21,000 4/15/2010 700 2,000 600:1 1,200,000
1.2m Deliveries per second
MEMORY HIERARCHY
Possible implementations • Fanout to disk • Ridonculous number of IOPS required, even with fancy buffering techniques • Cost of rebuilding data from other durable stores not too expensive • Fanout to memory • Good if cardinality of corpus * bytes/datum not too many GB
Low Latency get append fanout 1ms 1ms <1s* * Depends on the number of followers of the tweeter
Principles • Off-line vs. Online computation • The answer to some problems can be pre-computed if the amount of work is bounded and the query pattern is very limited • Keep the memory hierarchy in mind • The efficiency of a system includes the cost of generating data from another source (such as a backup) times the probability of needing to
The four data problems • Tweets • Timelines • Social graphs • Search indices
Big Data in Real-Time at Twitter
What is a Social Graph? • List of who follows whom, who blocks whom, etc. • Operations: • Enumerate by time • Intersection, Union, Difference • Inclusion • Cardinality • Mass-deletes for spam • Medium-velocity unbounded vectors • Complex, predetermined queries
Big Data in Real-Time at Twitter
Temporal enumeration
Inclusion Temporal enumeration
Inclusion Temporal enumeration Cardinality
Big Data in Real-Time at Twitter
Intersection: Deliver to people who follow both @aplusk and @foursquare
Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
Index Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
Index Index Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
Problems w/ solution • Write throughput • Indices couldn’t be kept in RAM
Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions Partitioned by user • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
Edges stored in both directions Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions Partitioned by user • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
Challenges • Data consistency in the presence of failures • Write operations are idempotent: retry until success • Last-Write Wins for edges • (with an ordering relation on State for time conflicts) • Other commutative strategies for mass-writes
Low Latency write cardinality iteration write ack inclusion materialize 1ms 100edges/ms* 1ms 16ms 1ms * 2ms lower bound
Principles • It is not possible to pre-compute set algebra queries • Simple distributed coordination techniques work • Partition, replicate, index. Many efficiency and scalability problems are solved the same way
The four data problems • Tweets • Timelines • Social graphs • Search indices
Big Data in Real-Time at Twitter
What is a Search Index? • “Find me all tweets with these words in it...” • Posting list • Boolean and/or queries • Complex, ad hoc queries • Relevance is recency* * Note: there is a non-real-time component to search, but it is not discussed here
Intersection of three posting lists
Original Implementation term_id doc_id 20 12 20 86 34 16 • Single table, vertically scaled • Master-Slave replication for read throughput
Problems w/ solution • Index could not be kept in memory
Current Implementation term_id doc_id 24 ... Partition 2 23 ... term_id doc_id Partition 1 22 ... 21 ... • Partitioned by time • Uses MySQL • Uses delayed key-write
Problems w/ solution • Write throughput • Queries for rare terms need to search many partitions • Space efficiency/recall • MySQL requires lots of memory
DATA NEAR COMPUTATION
Future solution • Document partitioning • Time partitioning too • Merge layer • May use Lucene instead of MySQL
Principles • Partition so that work can be parallelized • Temporal locality is not always enough
The four data problems • Tweets • Timelines • Social graphs • Search indices
Summary Statistics writes/ reads/second cardinality bytes/item durability second Tweets 100k 850 12b 300b durable Timelines 80k 1.2m a lot 3.2k non Graphs 100k 20k 13b 110 durable Search 13k 21k† 315m‡ 1k durable † tps * 25 postings ‡ 75 partitions * 4.2m tweets
Principles • All engineering solutions are transient • Nothing’s perfect but some solutions are good enough for a while • Scalability solutions aren’t magic. They involve partitioning, indexing, and replication • All data for real-time queries MUST be in memory. Disk is for writes only. • Some problems can be solved with pre-computation, but a lot can’t • Exploit locality where possible
Appendix

More Related Content

Big Data in Real-Time at Twitter

  • 1. Big Data in Real-Time at Twitter by Nick Kallen (@nk)
  • 3. What is Real-Time Data? • On-line queries for a single web request • Off-line computations with very low latency • Latency and throughput are equally important • Not talking about Hadoop and other high-latency, Big Data tools
  • 4. The four data problems • Tweets • Timelines • Social graphs • Search indices
  • 6. What is a Tweet? • 140 character message, plus some metadata • Query patterns: • by id • by author • (also @replies, but not discussed here) • Row Storage
  • 7. Find by primary key: 4376167936
  • 8. Find all by user_id: 749863
  • 9. Original Implementation id user_id text created_at 20 12 just setting up my twttr 2006-03-21 20:50:14 29 12 inviting coworkers 2006-03-21 21:02:56 34 16 Oh shit, I just twittered a little. 2006-03-21 21:08:09 • Relational • Single table, vertically scaled • Master-Slave replication and Memcached for read throughput.
  • 10. Original Implementation Master-Slave Replication Memcached for reads
  • 11. Problems w/ solution • Disk space: did not want to support disk arrays larger than 800GB • At 2,954,291,678 tweets, disk was over 90% utilized.
  • 13. Possible implementations Partition by primary key Partition 1 Partition 2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... 24 ... 25 ...
  • 14. Possible implementations Partition by primary key Partition 1 Partition 2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... 24 ... 25 ... Finding recent tweets by user_id queries N partitions
  • 15. Possible implementations Partition by user id Partition 1 Partition 2 id user_id id user_id ... 1 21 2 ... 1 23 2 ... 3 25 2
  • 16. Possible implementations Partition by user id Partition 1 Partition 2 id user_id id user_id ... 1 21 2 ... 1 23 2 ... 3 25 2 Finding a tweet by id queries N partitions
  • 17. Current Implementation Partition by time id user_id 24 ... Partition 2 23 ... id user_id Partition 1 22 ... 21 ...
  • 18. Current Implementation Queries try each Partition by time partition in order id until enough data user_id 24 ... is accumulated Partition 2 23 ... id user_id Partition 1 22 ... 21 ...
  • 20. Low Latency PK Lookup Memcached 1ms MySQL <10ms* * Depends on the number of partitions searched
  • 21. Principles • Partition and index • Exploit locality (in this case, temporal locality) • New tweets are requested most frequently, so usually only 1 partition is checked
  • 22. Problems w/ solution • Write throughput • Have encountered deadlocks in MySQL at crazy tweet velocity • Creating a new temporal shard is a manual process and takes too long; it involves setting up a parallel replication hierarchy. Our DBA hates us
  • 23. Future solution Partition k1 Partition k2 id user_id id user_id 20 ... 21 ... 22 ... 23 ... Partition u1 Partition u2 user_id ids user_id ids 12 20, 21, ... 13 48, 27, ... 14 25, 32, ... 15 23, 51, ... • Cassandra (non-relational) • Primary Key partitioning • Manual secondary index on user_id • Memcached for 90+% of reads
  • 24. The four data problems • Tweets • Timelines • Social graphs • Search indices
  • 26. What is a Timeline? • Sequence of tweet ids • Query pattern: get by user_id • Operations: • append • merge • truncate • High-velocity bounded vector • Space-based (in-place mutation)
  • 28. Original Implementation SELECT * FROM tweets WHERE user_id IN (SELECT source_id FROM followers WHERE destination_id = ?) ORDER BY created_at DESC LIMIT 20
  • 29. Original Implementation SELECT * FROM tweets WHERE user_id IN (SELECT source_id FROM followers WHERE destination_id = ?) ORDER BY created_at DESC LIMIT 20 Crazy slow if you have lots of friends or indices can’t be kept in RAM
  • 31. Current Implementation • Sequences stored in Memcached • Fanout off-line, but has a low latency SLA • Truncate at random intervals to ensure bounded length • On cache miss, merge user timelines
  • 32. Throughput Statistics date average tps peak tps fanout ratio deliveries 10/7/2008 30 120 175:1 21,000 4/15/2010 700 2,000 600:1 1,200,000
  • 35. Possible implementations • Fanout to disk • Ridonculous number of IOPS required, even with fancy buffering techniques • Cost of rebuilding data from other durable stores not too expensive • Fanout to memory • Good if cardinality of corpus * bytes/datum not too many GB
  • 36. Low Latency get append fanout 1ms 1ms <1s* * Depends on the number of followers of the tweeter
  • 37. Principles • Off-line vs. Online computation • The answer to some problems can be pre-computed if the amount of work is bounded and the query pattern is very limited • Keep the memory hierarchy in mind • The efficiency of a system includes the cost of generating data from another source (such as a backup) times the probability of needing to
  • 38. The four data problems • Tweets • Timelines • Social graphs • Search indices
  • 40. What is a Social Graph? • List of who follows whom, who blocks whom, etc. • Operations: • Enumerate by time • Intersection, Union, Difference • Inclusion • Cardinality • Mass-deletes for spam • Medium-velocity unbounded vectors • Complex, predetermined queries
  • 46. Intersection: Deliver to people who follow both @aplusk and @foursquare
  • 47. Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
  • 48. Index Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
  • 49. Index Index Original Implementation source_id destination_id 20 12 29 12 34 16 • Single table, vertically scaled • Master-Slave replication
  • 50. Problems w/ solution • Write throughput • Indices couldn’t be kept in RAM
  • 51. Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
  • 52. Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions Partitioned by user • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
  • 53. Edges stored in both directions Current solution Forward Backward source_id destination_id updated_at x destination_id source_id updated_at x 20 12 20:50:14 x 12 20 20:50:14 x 20 13 20:51:32 12 32 20:51:32 20 16 12 16 • Partitioned by user id • Edges stored in “forward” and “backward” directions Partitioned by user • Indexed by time • Indexed by element (for set algebra) • Denormalized cardinality
  • 54. Challenges • Data consistency in the presence of failures • Write operations are idempotent: retry until success • Last-Write Wins for edges • (with an ordering relation on State for time conflicts) • Other commutative strategies for mass-writes
  • 55. Low Latency write cardinality iteration write ack inclusion materialize 1ms 100edges/ms* 1ms 16ms 1ms * 2ms lower bound
  • 56. Principles • It is not possible to pre-compute set algebra queries • Simple distributed coordination techniques work • Partition, replicate, index. Many efficiency and scalability problems are solved the same way
  • 57. The four data problems • Tweets • Timelines • Social graphs • Search indices
  • 59. What is a Search Index? • “Find me all tweets with these words in it...” • Posting list • Boolean and/or queries • Complex, ad hoc queries • Relevance is recency* * Note: there is a non-real-time component to search, but it is not discussed here
  • 61. Original Implementation term_id doc_id 20 12 20 86 34 16 • Single table, vertically scaled • Master-Slave replication for read throughput
  • 62. Problems w/ solution • Index could not be kept in memory
  • 63. Current Implementation term_id doc_id 24 ... Partition 2 23 ... term_id doc_id Partition 1 22 ... 21 ... • Partitioned by time • Uses MySQL • Uses delayed key-write
  • 64. Problems w/ solution • Write throughput • Queries for rare terms need to search many partitions • Space efficiency/recall • MySQL requires lots of memory
  • 66. Future solution • Document partitioning • Time partitioning too • Merge layer • May use Lucene instead of MySQL
  • 67. Principles • Partition so that work can be parallelized • Temporal locality is not always enough
  • 68. The four data problems • Tweets • Timelines • Social graphs • Search indices
  • 69. Summary Statistics writes/ reads/second cardinality bytes/item durability second Tweets 100k 850 12b 300b durable Timelines 80k 1.2m a lot 3.2k non Graphs 100k 20k 13b 110 durable Search 13k 21k† 315m‡ 1k durable † tps * 25 postings ‡ 75 partitions * 4.2m tweets
  • 70. Principles • All engineering solutions are transient • Nothing’s perfect but some solutions are good enough for a while • Scalability solutions aren’t magic. They involve partitioning, indexing, and replication • All data for real-time queries MUST be in memory. Disk is for writes only. • Some problems can be solved with pre-computation, but a lot can’t • Exploit locality where possible

Editor's Notes