Graph parallel computation API; Graph processing library; Large-scale graph processing system Definition A graph processing framework (GPF) is a set of tools oriented to process graphs. Graph vertices are used to model data and edges model relationships between vertices. Typically, a GPF includes an input data stream, an execution model, and an application programming interface (API) having a set of functions implementing specific graph algorithms. In addition, some GPFs provide support for vertices and edges annotated with arbitrary properties of any kind and number. Graphs are being used for modeling large phenomena and their relationships in different application domains such as social network analysis (SNA), biological network analysis, and link analysis for fraud/cybercrime detection. Since real graphs can be large, complex, and dynamic, GPFs have to deal with the three challenges of data growth: volume, velocity, and variety. The programming API of GPFs is simple enough to facilitate efficient algorithm execution on graphs while abstracting away low-level details. On the one hand, the basic computation entity in a GPF can be graph-centric, vertexcentric, or edge-centric. On the other hand, as big data drives graph sizes beyond memory capacity of a single machine, GPFs partition and distribute data among computing nodes implementing a communication model. Overview There is a growing number of GPFs. Doekemeijer and Varbanescu (2014) review 83 different frameworks. To briefly explain the most relevant aspects of GPFs, 11 proposals, shown in Table 1, have been selected in this entry. Four main dimensions can be used to classify GPF frameworks: (1) the target platform, (2) the computation model, (3) the graph processing approach, and (4) the communication model. The target platform is one of the major distinctions between frameworks. Most of the GPFs are distributed memory frameworks (DMF), but there are some shared-memory frameworks (SMF) that exploit the parallel capabilities of a single machine.

No abstract