Evaluation of Parallel Graph Loading Techniques Manuel Then, Moritz - - PowerPoint PPT Presentation

▶
evaluation of parallel graph loading techniques
SMART_READER_LITE
LIVE PREVIEW

Evaluation of Parallel Graph Loading Techniques Manuel Then, Moritz - - PowerPoint PPT Presentation

May 17, 2023 229 likes •712 views

Evaluation of Parallel Graph Loading Techniques Manuel Then, Moritz Kaufmann, Alfons Kemper, Thomas Neumann Technical University of Munich Chair of Database Systems Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques 3

slide-1
SLIDE 1

Manuel Then, Moritz Kaufmann, Alfons Kemper, Thomas Neumann Technical University of Munich Chair of Database Systems

Evaluation of Parallel Graph Loading Techniques

slide-2
SLIDE 2

3 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

slide-3
SLIDE 3

4 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

slide-4
SLIDE 4

Goal: Efficiently load a given graph dataset for explorative analytics

5 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

slide-5
SLIDE 5

Problem: The optimal way of loading the graph depends on various factors:

  • Format of the graph data
  • Source of the data
  • Properties of the input data
  • Target graph data structure
  • Execution machine

Graph loading pipeline must be adapted to the scenario at hand

6 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Scenario-specific Graph Loading

slide-6
SLIDE 6

Goal: Efficiently load a given graph dataset for explorative analytics

7 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

slide-7
SLIDE 7

Goal: Efficiently load a given graph dataset for explorative analytics

8 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string?

slide-8
SLIDE 8

Goal: Efficiently load a given graph dataset for explorative analytics

9 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Can input data be read multiple times?

slide-9
SLIDE 9

Goal: Efficiently load a given graph dataset for explorative analytics

10 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Random access possible? Can input data be read multiple times?

slide-10
SLIDE 10

Goal: Efficiently load a given graph dataset for explorative analytics

11 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Random access possible? Can input data be read multiple times? Explicit vertex list available?

slide-11
SLIDE 11

Goal: Efficiently load a given graph dataset for explorative analytics

12 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Random access possible? Can input data be read multiple times? Explicit vertex list available?

slide-12
SLIDE 12

Goal: Efficiently load a given graph dataset for explorative analytics

13 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Random access possible? Can input data be read multiple times? Explicit vertex list available? Which data structure to generate?

slide-13
SLIDE 13

Goal: Efficiently load a given graph dataset for explorative analytics

14 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

Identifier data type? binary, decimal, string? Random access possible? Can input data be read multiple times? Explicit vertex list available? Which data structure to generate?

slide-14
SLIDE 14

Goal: Efficiently load a given graph dataset for explorative analytics

15 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

slide-15
SLIDE 15

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

16 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

slide-16
SLIDE 16

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

17 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

slide-17
SLIDE 17

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

18 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-18
SLIDE 18

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

19 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-19
SLIDE 19

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

20 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-20
SLIDE 20

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

21 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-21
SLIDE 21

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

Vectorized decimal parsing

  • Leverage wide vector units for identifier parsing

22 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-22
SLIDE 22

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

Vectorized decimal parsing

  • Leverage wide vector units for identifier parsing

23 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

2x 20x 200x

Parsers

  • T. Mủhlbauer, W. Rỏdiger, R. Seilbeck, A. Reiser, A. Kemper, and T. Neumann

Instant loading for main memory databases. Proceedings of the VLDB Endowment, 2013.

slide-23
SLIDE 23

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

Vectorized decimal parsing

  • Leverage wide vector units for identifier parsing

24 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-24
SLIDE 24

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

Vectorized decimal parsing

  • Leverage wide vector units for identifier parsing

25 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-25
SLIDE 25

Binary reader

  • No parsing necessary => directly copy vertex identifiers
  • Every edge same size => work splitting trivial

Library-provided decimal parsing

  • Readily-available for many languages
  • We evaluated C++’s stream operator and strtol
  • Varying edge length => work splitting more complex

Iterative decimal parsing

  • Multiply by ten and add character’s respective digit

Vectorized decimal parsing

  • Leverage wide vector units for identifier parsing

Parser code generation

26 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Parsers

2x 20x 200x

slide-26
SLIDE 26

Goal: Efficiently load a given graph dataset for explorative analytics

27 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

slide-27
SLIDE 27

Closely related areas

28 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Data Structures and Identifier Relabeling

slide-28
SLIDE 28

Closely related areas Map of Neighbor Lists => No relabeling (Identity)

  • Directly use dataset identifiers
  • Runtime overhead for neighbor and property accesses
  • Simple and efficient to load

29 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Data Structures and Identifier Relabeling

1 1 2 2

slide-29
SLIDE 29

Closely related areas Map of Neighbor Lists => No relabeling (Identity)

  • Directly use dataset identifiers
  • Runtime overhead for neighbor and property accesses
  • Simple and efficient to load

30 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Data Structures and Identifier Relabeling

1 1 2 2

Hash-based access

slide-30
SLIDE 30

Closely related areas Map of Neighbor Lists => No relabeling (Identity)

  • Directly use dataset identifiers
  • Runtime overhead for neighbor and property accesses
  • Simple and efficient to load

Compressed Sparse Row (CSR) => Dense relabeling

  • Dense identifiers [0, |V|-1]
  • Packed, sequential memory layout
  • Allows offset-based data structure access
  • e.g. for neighbor lists, or properties
  • Overhead during loading

31 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Data Structures and Identifier Relabeling

1 1 2 2

1 1 2 2 Hash-based access

slide-31
SLIDE 31

Closely related areas No relabeling (Identity) => Map of Neighbor Lists

  • Directly use dataset identifiers
  • Runtime overhead for neighbor and property accesses
  • Simple and efficient to load

Dense relabeling => Compressed Sparse Row (CSR)

  • Dense identifiers [0, |V|-1]
  • Packed, sequential memory layout
  • Allows offset-based data structure access
  • e.g. for neighbor lists, or properties
  • Overhead during loading

32 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Data Structures and Identifier Relabeling

1 1 2 2

1 1 2 2 Hash-based access Offset-based access

slide-32
SLIDE 32

Mapping

  • Assign dense identifiers while reading the input data
  • Global: All workers use a shared map
  • Local: Each worker creates a local relabeling

33 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Relabeling Strategies

slide-33
SLIDE 33

Mapping

  • Assign dense identifiers while reading the input data
  • Global: All workers use a shared map
  • Local: Each worker creates a local relabeling

Collection

  • Gather unique identifiers while reading the input
  • Assign dense identifiers at the end
  • Global: Shared identifier set for all workers
  • Local: Use a local set per worker

34 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Relabeling Strategies

∪ ∪ ∪

slide-34
SLIDE 34

Mapping

  • Assign dense identifiers while reading the input data
  • Global: All workers use a shared map
  • Local: Each worker creates a local relabeling

Collection

  • Gather unique identifiers while reading the input
  • Assign dense identifiers at the end
  • Global: Shared identifier set for all workers
  • Local: Use a local set per worker

Relabeling is finalized/applied when the graph data structure is written

35 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Relabeling Strategies

∪ ∪ ∪

slide-35
SLIDE 35

Graph loading times for various relabeling strategies No further dataset properties leveraged

36 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Relabeling Strategies - Measurements

slide-36
SLIDE 36

Graph loading times for various relabeling strategies No further dataset properties leveraged

37 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Relabeling Strategies - Measurements

slide-37
SLIDE 37

Goal: Efficiently load a given graph dataset for explorative analytics

38 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

General Graph Loading Pipeline

Read

  • Parse edges and create relabeling
  • Write edges to worker-local buffer

Sync

  • Find unique vertices
  • Count neighbors

Write

  • Create final graph data structure
  • Apply final relabeling

Analytics • The actual analytics work

slide-38
SLIDE 38

Explicit vertex lists

  • All unique vertices in the dataset are known beforehand
  • No need to find and count vertices => improves loading efficiency

39 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties

slide-39
SLIDE 39

Explicit vertex lists

  • All unique vertices in the dataset are known beforehand
  • No need to find and count vertices => improves loading efficiency

Partitioned edge list

  • Edge list partitioned by source vertex
  • Each source vertex has a responsible worker thread
  • determined by the input data chunk
  • Significantly reduces worker communication overhead

40 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties

slide-40
SLIDE 40

Explicit vertex lists

  • All unique vertices in the dataset are known beforehand
  • No need to find and count vertices => improves loading efficiency

Partitioned edge list

  • Edge list partitioned by source vertex
  • Each source vertex has a responsible worker thread
  • determined by the input data chunk
  • Significantly reduces worker communication overhead

41 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties

Partitioned 1 2 1 3 1 4 2 1 2 4 3 1 3 2 4 3 Unpartitioned 4 3 1 3 3 1 1 4 2 1 1 2 3 2 2 4

slide-41
SLIDE 41

42 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties - Measurements

Graphs

  • LDBC-1000, |V| = 3.6M, |E| = 447M
  • Twitter , |V| = 41.6M, |E| = 1.5B
slide-42
SLIDE 42

43 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties - Measurements

Graphs

  • LDBC-1000, |V| = 3.6M, |E| = 447M
  • Twitter , |V| = 41.6M, |E| = 1.5B
slide-43
SLIDE 43

44 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties - Measurements

Graphs

  • LDBC-1000, |V| = 3.6M, |E| = 447M
  • Twitter , |V| = 41.6M, |E| = 1.5B
slide-44
SLIDE 44

45 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Leveraging Dataset Properties - Measurements

Graphs

  • LDBC-1000, |V| = 3.6M, |E| = 447M
  • Twitter , |V| = 41.6M, |E| = 1.5B
slide-45
SLIDE 45

46 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Comparison with Existing Systems

Twitter LDBC Oracle PGX 2153s 632s GraphBIG

  • ut of memory

1682s Ours non-partitioned 88s 24s Ours partitioned 34s 7s

Graphs

  • LDBC-1000, |V| = 3.6M, |E| = 447M
  • Twitter , |V| = 41.6M, |E| = 1.5B

Machine:

  • 2x Intel Xeon E5-2660 v2 2 × 20 @ 2.2GHz)
  • 256GB, Ubuntu 15.10, kernel 4.2.0
slide-46
SLIDE 46

CSR (relabeled) Load + Run = Total Neighbors Map (identity) Load + Run = Total PageRank 37s 33s 70s---- 25s 194s 219s---- Triangle Counting 37s 49s 86s---- 25s 66s 92s----

47 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Influence on Analytics

Graphs

  • Twitter , |V| = 41.6M, |E| = 1.5B

Machine:

  • 2x Intel Xeon E5-2660 v2 2 × 20 @ 2.2GHz)
  • 256GB, Ubuntu 15.10, kernel 4.2.0
slide-47
SLIDE 47

Optimal loading pipeline for a graph dataset is highly dependent on the

  • Data format
  • Source of the data
  • Properties of the dataset
  • Algorithm-dependent graph data structure
  • Target machine

Custom iterative identifier parsing always beneficial Concurrent identifier relabeling mostly beneficial

  • More challenging than identity mapping, but usually worth it

Leveraging properties of the dataset can lead to enormous speedups

48 Manuel Then (TUM) | Evaluation of Parallel Graph Loading Techniques

Summary