High-performance analysis of filtered semantic graphs

  • Authors:

  • Aydin Buluç;Armando Fox;John R. Gilbert;Shoaib A. Kamil;Adam Lugowski;Leonid Oliker;Samuel W. Williams

  • Affiliations:

  • Lawrence Berkeley National Laboratory, Berkeley, USA;University of California, Berkeley, Berkeley, USA;University of California, Santa Barbara, Santa Barbara, USA;University of California, Berkeley, Berkeley, USA;University of California, Santa Barbara, Santa Barbara, USA;Lawrence Berkeley National Laboratory, Berkeley, USA;Lawrence Berkeley National Laboratory, Berkeley, USA

  • Venue:
  • Proceedings of the 21st international conference on Parallel architectures and compilation techniques
  • Year:
  • 2012

Quantified Score

Hi-index 0.00

Visualization

Abstract

High performance is a crucial consideration when executing a complex analytic query on a massive semantic graph. In a semantic graph, vertices and edges carry "attributes" of various types. Analytic queries on semantic graphs typically depend on the values of these attributes; thus, the computation must either view the graph through a "filter" that passes only those individual vertices and edges of interest, or else must first materialize a subgraph or subgraphs consisting of only the vertices and edges of interest. The filtered approach is superior due to its generality, ease of use, and memory efficiency, but may carry a performance cost. In the Knowledge Discovery Toolbox (KDT), a Python library for parallel graph computations, the user writes filters in a high-level language, but those filters result in relatively low performance due to the bottleneck of having to call into the Python interpreter for each edge. In this work, we use the Selective Embedded JIT Specialization (SEJITS) approach to automatically translate filters defined by programmers into a lower-level efficiency language, bypassing the upcall into Python. We evaluate our approach by comparing it with the high-performance C++/MPI Combinatorial BLAS engine, and show that the productivity gained by using a high-level filtering language comes without sacrificing performance.