The document discusses popular algorithms for storing data on disk, including B-trees and LSM-trees. B-trees balance speed for insert, update, and select operations, though deletes can be slow. LSM-trees are suitable for data with varying recency where older data can be aggregated, like messages in chat logs or social networks. LSM-trees segregate data between memory and disk for indexing.

1. Популярные алгоритмы
хранения данных на диске
Konstantin Osipov,
kostja@tarantool.org
October 28th, 2013

2. Случай в квадрате 36-80
•
•
•
B-tree – most
popular disk-based
data structure
B-tree balances
INSERT, UPDATE
and SELECT speed
DELETEs can be
slow

3. СУБД быстрая, настраивать
надо уметь

4. B-tree: внутреннее устройство

5. Что означает сache-oblivious?

6. Что означает сache-oblivious? (2)
BLOCK­MULT(A,B,C,n):
1 for i = 1 to n/s do:
2 for j = 1 to n/s do:
3 for k = 1 to n/s do:
4 ORD­MULT(Aik, Bkj, Cij, s)

7. LSM-tree: внутреннее
устройство

8. LSM-tree: внутреннее
устройство (2)

9. LevelDB: устройство

10. LevelDB: insert RPS

11. LSM-tree: применение
●
Данные с разной степенью актуальности
–
–
Стена в соцсети
–
Чаты
–
●
Ленты сообщений
События
Сегрегация данных
–
Данные в LSM space, индекс в MEMORY space

12. TokuDB/CO lookahead arrays

13. PUT(37), PUT(16)
Self-Balancing Tree
Memory
Disk
WAL:

14. 16 37 Self-Balancing Tree Memory
Disk
WAL: 37, 16

15. 7 41 Self-Balancing Tree Memory
16 37 Sorted String Table
WAL: 41, 7, 37, 16
Disk

16. Memory
7 37
7 16 37 41
WAL: 41, 7, 28, 16
Disk

17. 10 28
Memory
7 37
Disk
7 16 37 41
WAL: 10, 28, 41, 7, 37, 16

18. Memory
10 28
7 16 37 41
WAL: 10, 28, 41, 7, 37, 16
Disk

19. 2 47
Memory
10 28
Disk
7 16 37 41
WAL: 47, 2, 10, 28, 41, 7, 37, 16

20. 2 28
2 10 28 41
2
7 10 16 28 37 41 47
WAL: 47, 2, 10, 28, 41, 7, 37, 16
Memory
Disk

21. 6 49
2 28
2 10 28 41
2
7 10 16 28 37 41 47
WAL: 49, 6, 47, 2, 10, 28, 41, 7, 37, 16
Memory
Disk

22. 6 49
2 10 28 41
2
7 10 16 28 37 41 47
WAL: 49, 6, 47, 2, 10, 28, 41, 7, 37, 16
Memory
Disk

23. 23 32
6 49
2 10 28 41
2
7 10 16 28 37 41 47
WAL: 32, 23, 49, 6, 47, 2, 10, 28, 41, 7, 37, 16
Memory
Disk

24. 6 32
6 23 32 49
2
7 10 16 28 37 41 47
WAL: 32, 23, 49, 6, 47, 2, 10, 28, 41, 7, 37, 16
Memory
Disk

25. 30 45
Memory
6 32
Disk
6 23 32 49
2
7 10 16 28 37 41 47
WAL: 30, 45, 32, 23, 49, 6, 47, 2, 10, 28, 41, 7, 37, 16

26. 14 38
Memory
30 45
Disk
6 23 32 49
2
7 10 16 28 37 41 47
WAL: 38, 14, 30, 45, 32, 23, 49, 6, 47, 2, 10, 28, 41, 7, 37, 16

27. 6 10
Memory
2 30
Disk
2 14 30 41
2
2
6
7 14 23 30 37 41 47
7 10 14 16 23 28 30 32 37 38 41 45 47 49
WAL: 10, 6, 38, 14, 45, 30, 45, 32, 23, 49, 6, 47, 2, 10, 28, 41, 7, 37, 16

28. Memory
22 37
Disk
10 25 36 42
3
2
6
8 15 26 35 40 45 48
7 10 14 16 23 28 30 32 37 38 41 45 47 49
WAL: 37, 22, 36, 10, 25, 42, 10, 6, 38, 14, 45, 30, 45, 32, 23, 49, 6, 47, 2,
10, 28, 41, 7, 37, 16

29. GET(16)
Memory
22 37
Disk
10 25 36 42
3
2
6
8 15 26 35 40 45 48
7 10 14 16 23 28 30 32 37 38 41 45 47 49
WAL: 37, 22, 36, 10, 25, 42, 10, 6, 38, 14, 45, 30, 45, 32, 23, 49, 6, 47, 2,
10, 28, 41, 7, 37, 16

30. GET(16)
Memory
22 37
Disk
10 25 36 42
3
2
6
8 15 26 35 40 45 48
7 10 14 16 23 28 30 32 37 38 41 45 47 49
WAL: 37, 22, 36, 10, 25, 42, 10, 6, 38, 14, 45, 30, 45, 32, 23, 49, 6, 47, 2,
10, 28, 41, 7, 37, 16

31. BitCask: AOF format

32. BitCask: key dir

33. Sophia:

34. Links
●
●
●
●
●
●
●
●
●
●
Bitcask A Log-Structured Hash Table for Fast Key/Value Data, Justin Sheehy David Smith with
inspiration from Eric Brewer
The Log-Structured Merge-Tree (LSM-Tree) Patrick O'Neil , Edward Cheng, Dieter Gawlick,
Elizabeth O'Neil
Cache-Oblivious Algorithms by Harald Prokop (Master theses)
Space/time trade-offs in hash coding with allowable errors, Burton H. Bloom
Data Structures and Algorithms for Big Databases, Michael A. Bender Stony Brook & Tokutek
Bradley C. Kuszmaul (XLDB tutorial)
http://github.com/pmwkaa/sophia, http://sphia.org
http://codecapsule.com/2012/12/30/implementing-a-key-value-store-part-3-comparative-analysis-of-the-ar
http://stackoverflow.com/questions/6079890/cache-oblivious-lookahead-array
http://www.youtube.com/watch?v=88NaRUdoWZM(Tim Callaghan: Fractal Tree indexes)
http://code.google.com/p/leveldb/downloads/list

35. Эпилог: choose your db wisely
?