Giters

HybridGraph / HGraph

HGraph(HybridGraph) is a Pregel-like system which merges Pulling/Pushing for I/O-Efficient distributed and iterative graph computing.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

News!!!
[10/30/2019] A new paper is accepted by IEEE TKDE!!!
[12/10/2018] We change the project name from HybridGraph to HGraph for short!!!
[11/20/2018] Priority scheduling for BPull is now available.
[10/20/2018] More example graph algorithms are packaged.
[09/09/2018] A general switching framework is added.
[05/11/2017] Lightweight fault-tolerance is added.
[04/05/2017] Some engineering optimizations are added for better performance.
[02/17/2017] Some APIs are changed.
[12/25/2016] Code cleanup.
[05/15/2016] HybridGraph is online.

HGraph

HGraph(HybridGraph) is a Pregel-like system which merges Pulling/Pushing for I/O-Efficient distributed and iterative graph computing.

1. Introduction

Billion-scale graphs are rapidly growing in size in many applications. That has been driving much of the research on enhancing the performance of distributed graph systems, including graph partitioning, network communication, and the convergence. HGraph focuses on performing graph analysis on a cluster I/O-efficiently, since the memory resource of a given computer cluster can be easily exhausted. Specially, HGraph employs a hybrid solution to support switching between push and pull adaptively to obtain optimal performance in different scenarios.

The HGraph project started at Northeastern University (China) in 2011. HGraph is a Java framework implemented on top of Apache Hama 0.2.0-incubating. Features of HGraph are listed as below.

  • Block-centric pull mechanism (BPull): I/O accesses are shifted from receiver sides where messages are written/read by Push to sender sides where graph data are read by Pull. The cost of random reads regarding vertices in existing pull-based approaches is considerable. Thus, a novel block-centric pull technique is used to optimize the I/O-efficiency, as well as reducing the communication cost caused by sending pull requests.
  • VE-BLOCK storage: A disk-resident block-centric graph structure is designed for efficient data accesses in BPull. Graph data are separated into vertices and edges. Vertices are divided into several VBlocks. Accordingly, edges are partitioned into multiple EBlocks. Messages are pulled in VBlocks.
  • Hybrid engine: By analyzing behaviors of different algorithms, two seamless switching mechanisms as well as prominent performance prediction methods are proposed to guarantee the efficiency when switching between Push and BPull.
  • Lightweight fault-tolerance: By utilizing the features of BPull and the swiching mechanism, a new fault-tolerant framework is proposed to preform an efficient confined recovery without logging messages. It strikes a good balance between recovery efficiency and failure-free performance.
  • Priority scheduling: When using BPull, users can specify the parallelism of pull-responding to limit the resource consumption. HGraph can smartly schedule the responding order to minimize the responding time.
  • Algorithms library: PageRank, Shortest Path, Connected Components, Label Propagation, Maximal Independent Sets, and Bipartite Matching.  

2. Quick Start

This section describes how to configurate, compile and then deploy HGraph on a cluster consisting of three physical machines running Red Hat Enterprise Linux 6.4 32/64 bit (one master and two slaves/workers, called master, slave1, and slave2). Before that, Apache Hadoop should be installed on the cluster, which is beyond the scope of this document.

2.1 Requirements

  • Apache hadoop-0.20.2 (distributed storage service)
  • Sun Java JDK 1.6.x or higher version
    Without loss of generality, suppose that HGraph is installed in ~/HGraph and Java is installed in /usr/java/jdk1.6.0_23.

2.2 Deploying HGraph

downloading files on master

cd ~/
git clone https://github.com/HybridGraph/HGraph.git
chmod 777 -R HGraph/

configuration on master

First, edit /etc/profile by typing sudo vi /etc/profile and then add the following information:
export HGraph_HOME=~/HGraph
export HGraph_CONF_DIR=$HGraph_HOME/conf
export PATH=$PATH:$HGraph_HOME/bin
After that, type source /etc/profile in the command line to make changes take effect.

Second, edit configuration files in HGraph_HOME/conf as follows:

  • termite-env.sh: setting up the Java path.
    export JAVA_HOME=/usr/java/jdk1.6.0_23
  • termite-site.xml: configurating the HGraph engine.
    The details are shown in termite-site.xml.
    In particular, our current implementation uses the read/write throughput of disk reported by the disk benchmarking tool fio-2.0.13, and the network throughput reported by the network benchmarking tool iperf-2.0.5.
  1. random read/write
    fio -filename=/tmp/data -direct=1 -iodepth 1 -thread -rw=randrw -rwmixread=50 -ioengine=psync -bs=16k -size=10G -numjobs=30 -runtime=360 -group_reporting -name=test
  2. sequential read/write
    fio -filename=/tmp/data -direct=1 -iodepth 1 -thread -rw=rw -rwmixread=50 -ioengine=psync -bs=16k -size=10G -numjobs=30 -runtime=360 -group_reporting -name=test
  3. network
    server: iperf -s -f M -w 48K
    client: iperf -c hadoop03 -f M -w 48K
  • workers: settting up workers of HGraph.
    slave1
    slave2

deploying

Copy configurated files on master to slave1 and slave2.
scp -r ~/HGraph slave1:.
scp -r ~/HGraph slave2:.

2.3 Starting HGraph

Type the following commands on master to start HGraph.

  • starting HDFS:
    start-dfs.sh
  • starting HGraph after NameNode has left safemode:
    start-termite.sh
  • stopping HGraph:
    stop-termite.sh

2.4 Running a Single Source Shortest Path (SSSP) job on master

First, create an input file under input/random_graph on HDFS. Input file should be in format of:
source_vertex_id \t target_vertex_id_1:target_vertex_id_2:...
An example is given in random_graph. You can download it and put it onto your HDFS:
hadoop dfs -mkdir input
hadoop dfs -put random_graph input/
Currently, HGraph uses Range (a simple variant of Range used in Giraph) to partition input graph, in order to preserve the locality of raw graph. As a negative result of that, vertex ids must be numbered consecutively.

Second, submit the SSSP job with different models for the example graph random_graph:

  • SSSP (using b-pull):
    termite jar $HGraph_HOME/termite-examples-0.1.jar sssp.pull input output 2 50 100000 5 10000 2
    About arguments:
    [1] input directory on HDFS
    [2] output directory on HDFS
    [3] the number of child processes (tasks)
    [4] the maximum number of supersteps
    [5] the total number of vertices
    [6] the number of VBlocks per task
    [7] the sending threshold
    [8] the source vertex id
  • SSSP (using hybrid):
    termite jar $HGraph_HOME/termite-examples-0.1.jar sssp.hybrid input output 2 50 100000 5 10000 10000 10000 2 2
    About arguments:
    [1] input directory on HDFS
    [2] output directory on HDFS
    [3] the number of child processes (tasks)
    [4] the maximum number of supersteps
    [5] the total number of vertices
    [6] the number of VBlocks per task
    [7] the sending threshold used by b-pull
    [8] the sending threshold used by push
    [9] the receiving buffer size per task used by push
    [10] starting style: 1--push, 2--b-pull
    [11] the source vertex id

HGraph manages graph data on disk as default. Users can tell HGraph to keep graph data in memory through BSPJob.setGraphDataOnDisk(false). Currently, the memory version only works for BPull. Please type termite jar $HGraph_HOME/termite-examples-0.1.jar to list all example algorithms. A chosen algorithm will print usage help when no arguments is given.

3 Building HGraph with Apache Ant

Users can import source code into Eclipse as an existing Java project to modify the core engine of HGraph, and then build your modified version. Before building, you should install Apache Ant 1.7.1 or higher version on your master. Suppose the modified version is located in ~/source/HGraph. You can build it using ~/source/HGraph/build.xml as follows:
cd ~/source/HGraph
ant
Notice that you can build a specified part of HGraph as follows:

  1. build the core engine
    ant core.jar
  2. build examples
    ant examples.jar

By default, all parts will be built, and you can find termite-core-0.1.jar and termite-examples-0.1.jar in ~/source/HGraph/build after a successful building. Finally, use the new termite-core-0.1.jar to replace the old one in $HGraph_HOME on the cluster (i.e., master, slave1, and slave2). At anytime, you should guarantee that the termite-core-xx.jar file is unique in $HGraph_HOME. Otherwise, the starting script described in Section 2.3 may use a wrong file to start the HGraph engine.

4. Programming Guide

HGraph includes some simple graph algorithms to show the usage of its APIs. These algorithms are contained in the src/examples/hybrid/examples package and have been packaged into the termite-examples-0.1.jar file. Users can implement their own algorithms by learning these examples. After that, as described in Section 3 and Section 2.4, you can build your own algorithm can then run it.

6. Publications

7. Contact

If you encounter any problem with HGraph, please feel free to contact wangzhigang@ouc.edu.cn.

About

HGraph(HybridGraph) is a Pregel-like system which merges Pulling/Pushing for I/O-Efficient distributed and iterative graph computing.

Languages

Language:Shell 100.0%