Evaluation of embeddable graph manipulation libraries in memory constrained environments

Abstract

Memory leaks are major problems in distributed applications, depleting their performance, even if they run on platforms with automatic memory management like Java Virtual Machine. In the context of a Java memory leak detection tool, we faced a need for a graph storage and traversal library that could operate in memory restrained conditions with certain performance requirements. Before implementing a custom library we have evaluated existing open source graph libraries like JGraphT, Jung, jdsl, hgdb, Annas, HyperGraphDB, Dex and Neo4j, that can be used by embedding rather than a separate graph database process. Criteria for the comparison include size of the graph on disk, cpu and memory requirement for performing the shortest path search, license agreement terms. From this evaluation it is observed that graph databases become ideal choice only when hardware can be scaled for large problems, making it necessary for tailored solutions for managing large graphs in embedded environment. Copyright 2012 ACM.