Streaming Graph partitioning

This repository contains the slightly modified version of Fennel, and Linear Deterministic Greedy (LDG) partitioning algorithms to work with stream of edges (a pair of vertices). The only difference is , Practitioner accept edge of a graph instead of single vertexes and add both vertexes(aka Nodes) to partitions.

I have develp this code for testing out different streaming graph partitioning algorithms, and validate and try out the partioninig quality with visualization.

Technology stack

Partitioning algorithms are implemented in cpp+11 and visualization is done using Cytoscape.js. Use Apache Kafka for consume stream of edges in a graph. A python script is used for streaming edges from a graph in a file. It can also stream a graph by taking user inputs from the terminal. Practitioner accept the graph as a stream of edges , where edge contains a space separated vertex pair, i:e 22 45

Running

Installing

This tool uses cmake for managing package versions and building the binaries. CMake 3.10 or latest version is required to build the tool.

Moreover following packages are being used for the purpose mentioned below. You need to install them seperatly.

• Pistache : Use for developing the websocket and REST API for the live streaming and fetching graph meta data

• GCC 9.1.0 (recommended) or above

• SQLite3 (https://www.sqlite.org/download.html)

  • Download the sqlite-autoconf-3390200 from this URL.
  • Extract sqlite-autoconf-3390200.tar.gz to some location. E.g., /media/user/software/sqlite-autoconf-3390200-install
  • Change to that location and run ./configure --prefix=/media/user/software/sqlite-autoconf-3390200
  • Issue "make" followed by "sudo make install"
  • Once installed specify the target_link_libraries path to libsqlite3. E.g., target_link_libraries(JasmineGraph /media/user/software/sqlite-autoconf-3390200/lib/libsqlite3.so)

• cppkafka (https://github.com/mfontanini/cppkafka)

  • Install librdkafka - Follow the Readme in (https://github.com/edenhill/librdkafka) - use sudo apt install librdkafka-dev
  • Install boost library - use sudo apt-get install libboost-all-dev
  • Follow the guidelines in (https://github.com/mfontanini/cppkafka#compiling) - use cppkafka release v0.3.1
  • When doing the above step for cmake <OPTIONS> .. use of cmake .. should be sufficient enough
  • Once cppkafka is built install it by running sudo make install from the build directory

• SpdLog (https://github.com/gabime/spdlog)

  • Clone or download the repository from the above link
  • Issue "cmake ."
  • Issue "make" followed by "sudo make install"

• SQLite3 (https://www.sqlite.org/download.html)

  • Download the sqlite-autoconf-3390200 from this URL.
  • Extract sqlite-autoconf-3390200.tar.gz to some location. E.g., /media/user/software/sqlite-autoconf-3390200-install
  • Change to that location and run ./configure --prefix=/media/user/software/sqlite-autoconf-3390200
  • Issue "make" followed by "sudo make install"
  • Once installed specify the target_link_libraries path to libsqlite3. E.g., target_link_libraries(JasmineGraph /media/user/software/sqlite-autoconf-3390200/lib/libsqlite3.so)

• SQLite3 (https://www.sqlite.org/download.html)

  • Download the sqlite-autoconf-3390200 from this URL.
  • Extract sqlite-autoconf-3390200.tar.gz to some location. E.g., /media/user/software/sqlite-autoconf-3390200-install
  • Change to that location and run ./configure --prefix=/media/user/software/sqlite-autoconf-3390200
  • Issue "make" followed by "sudo make install"
  • Once installed specify the target_link_libraries path to libsqlite3. E.g., target_link_libraries(JasmineGraph /media/user/software/sqlite-autoconf-3390200/lib/libsqlite3.so)

• SQLite3 (https://www.sqlite.org/download.html)

  • Download the sqlite-autoconf-3390200 from this URL.
  • Extract sqlite-autoconf-3390200.tar.gz to some location. E.g., /media/user/software/sqlite-autoconf-3390200-install
  • Change to that location and run ./configure --prefix=/media/user/software/sqlite-autoconf-3390200
  • Issue "make" followed by "sudo make install"
  • Once installed specify the target_link_libraries path to libsqlite3. E.g., target_link_libraries(JasmineGraph /media/user/software/sqlite-autoconf-3390200/lib/libsqlite3.so)

• SQLite3 (https://www.sqlite.org/download.html)

  • Download the sqlite-autoconf-3390200 from this URL.
  • Extract sqlite-autoconf-3390200.tar.gz to some location. E.g., /media/user/software/sqlite-autoconf-3390200-install
  • Change to that location and run ./configure --prefix=/media/user/software/sqlite-autoconf-3390200
  • Issue "make" followed by "sudo make install"
  • Once installed specify the target_link_libraries path to libsqlite3. E.g., target_link_libraries(JasmineGraph /media/user/software/sqlite-autoconf-3390200/lib/libsqlite3.so)

More on Streaming Graphs

This was implemented as a part of adding new streaming graph partitioning algorithm to Jasmine Graph server. You could try out Jasmine graph server for more robust work on graph partitioning and analysis.

Algorithms

Implementation of Fennel

I have used following maximization formulae for deriving the partition index for most optimum partitioning. The hubristic used in here is, if a vertex having more neighbors and have not reached to it's maximum capacity, Then it(Partition) will have most likelihood to accept that vertex.

Inter partition cost

Intra partition cost

Results comparison

LDG

 ***Received the end of stream
0 => Vertext count = 2237
0 => Edges count = 1125
0 => Edge cuts count = 440
0 => Cut ratio = 0.28115
1 => Vertext count = 2235
1 => Edges count = 1162
1 => Edge cuts count = 430
1 => Cut ratio = 0.270101
2 => Vertext count = 2236
2 => Edges count = 1156
2 => Edge cuts count = 447
2 => Cut ratio = 0.278852
3 => Vertext count = 2236
3 => Edges count = 1153
3 => Edge cuts count = 455
3 => Cut ratio = 0.28296
Time taken = 3.88941e+07 micro seconds

Fennel

 ***Received the end of stream
0 => Vertext count = 1711
0 => Edges count = 688
0 => Edge cuts count = 1758
0 => Cut ratio = 0.718724
1 => Vertext count = 1711
1 => Edges count = 580
1 => Edge cuts count = 1900
1 => Cut ratio = 0.766129
2 => Vertext count = 1710
2 => Edges count = 553
2 => Edge cuts count = 1973
2 => Cut ratio = 0.781077
3 => Vertext count = 1711
3 => Edges count = 545
3 => Edge cuts count = 1961
3 => Cut ratio = 0.782522
Time taken = 5.8881e+07 micro seconds

Hash

 ***Received the end of stream
0 => Vertext count = 1644
0 => Edges count = 323
0 => Edge cuts count = 2601
0 => Cut ratio = 0.889535
1 => Vertext count = 1695
1 => Edges count = 310
1 => Edge cuts count = 2644
1 => Cut ratio = 0.895058
2 => Vertext count = 1672
2 => Edges count = 306
2 => Edge cuts count = 2648
2 => Cut ratio = 0.896412
3 => Vertext count = 1646
3 => Edges count = 326
3 => Edge cuts count = 2535
3 => Cut ratio = 0.886054
Time taken = 428249 micro seconds

Sample output

 ***Received the end of stream
0 => Vertext count = 5
0 => Edges count = 4
0 => Edge cuts count = 4
0 => Cut ratio = 0.5
Printing edge cuts of 0 partition
8 ____
        | ---> 9


39 ____
        | ---> 36


14 ____
        | ---> 17


35 ____
        | ---> 34


Printing edge list of 0 partition
6 ____
        | ===> 8


7 ____
        | ===> 8




14 ____
        | ===> 16
        | ===> 29






1 => Vertext count = 5
1 => Edges count = 3
1 => Edge cuts count = 6
1 => Cut ratio = 0.666667
Printing edge cuts of 1 partition
9 ____
        | ---> 8


36 ____
        | ---> 39


19 ____
        | ---> 20


36 ____
        | ---> 37


14 ____
        | ---> 13


36 ____
        | ---> 38


Printing edge list of 1 partition
18 ____
        | ===> 19




35 ____
        | ===> 36




41 ____
        | ===> 42




2 => Vertext count = 4
2 => Edges count = 1
2 => Edge cuts count = 3
2 => Cut ratio = 0.75
Printing edge cuts of 2 partition
20 ____
        | ---> 19


37 ____
        | ---> 36


15 ____
        | ---> 13


Printing edge list of 2 partition
9 ____
        | ===> 10




3 => Vertext count = 4.5
3 => Edges count = 3
3 => Edge cuts count = 5
3 => Cut ratio = 0.625
Printing edge cuts of 3 partition
17 ____
        | ---> 14


34 ____
        | ---> 35


13 ____
        | ---> 14


38 ____
        | ---> 36


13 ____
        | ---> 15


Printing edge list of 3 partition
5 ____
        | ===> 13


12 ____
        | ===> 13




33 ____
        | ===> 34




Time taken = 28433 micro seconds