Cassovary is a simple "big graph" processing library for the JVM. Most JVM-hosted graph libraries are flexible but not space efficient. Cassovary is designed from the ground up to first be able to efficiently handle graphs with billions of nodes and edges. A typical example usage is to do large scale graph mining and analysis of a big network. Cassovary is written in Scala and can be used with any JVM-hosted language. It comes with some common data structures and algorithms.

Please follow the cassovary project on twitter at @cassovary for updates.

There is a subproject included here called cassovary-examples containing simple java and scala examples of using the library. See this README to get started with the examples.

Some other subprojects to check are cassovary-benchmarks for helping benchmark some graph algorithms and cassovary-server that exposes Cassovary on a web server.

Cassovary is built using sbt and was tested last using sbt version 0.13.

1. ./sbt update (might take a couple of minutes)
2. ./sbt test
3. ./sbt package

1. ./sbt publish-local
2. cd ../<dependent project>
3. ./sbt update

Cassovary is published to maven central with crosspath scala versions 2.9.3 and 2.10.x. To use with sbt, use:

libraryDependencies += "com.twitter" %% "cassovary" % "3.2.0"

The only dependency that Cassovary uses which is not bundled with it because of its size is it.unimi.dsi.fastutil. You can add that dependency in your sbt project as follows:

libraryDependencies += "it.unimi.dsi" % "fastutil" % "6.4.4"

There are many excellent graph mining libraries already in existence. Most of them have one or more of the following characteristics:

1. Written in C/C++. Examples include SNAP from Stanford and GraphLab from CMU. The typical way to use these from JVM is to use JNI bridges.
2. Sacrifice storage efficiency for flexibility. Examples include JUNG which is written in Java but stores nodes and edges as big objects.
3. Are meant to do much more, typically a full graph database. Examples include Neo4J.

On the other hand, Cassovary is intended to be easy to use in a JVM-hosted environment and yet be efficient enough to scale to billions of edges. It is deliberately not designed to provide any persistence or database functionality. Also, it currently skips any concerns of partitioning the graph and hence is not directly comparable to distributed graph processing systems like Apache Giraph. This allows complex algorithms to be run on the graph efficiently, an otherwise recurring issue with distributed graph processing systems because of the known difficulty of achieving good graph partitions. On the flip side, the size of the graph it works with is bounded by the memory available in a machine, though the use of space efficient data structures does not seem to make this a limitation for most practical graphs. For example, an ArrayBasedDirectedGraph instance of a unidirectional graph with 10M nodes and 1B edges consumes less than 6GB of memory, and scales linearly beyond that.

http://groups.google.com/group/twitter-cassovary

Please follow the cassovary project on twitter at @cassovary for updates.

Please report any bugs to: https://github.com/twitter/cassovary/issues

Thanks to all the contributors of Cassovary.

We use the Yourkit Java Profiler for profiling and tuning Cassovary.

Copyright 2014 Twitter, Inc.

Licensed under the Apache License, Version 2.0: http://www.apache.org/licenses/LICENSE-2.0