conclusion.tex

\section{Conclusion}
\label{sec:conclusion}

In this paper we introduced the LAGraph library, the rationale behind its design,  
and a performance baseline with the GAP benchmark suite.   We also introduced
a notation for graph algorithms expressed in terms of linear algebra which we hope becomes
a consensus-notation adopted by the 
larger ``Graphs as Linear Algebra'' community.

This paper defines the foundation for our future work on the LAGraph project.  
We plan to explore Python wrappers for LAGraph that work well for data analytics workflows.  
In addition to the GAP benchmark, which focuses on graph algorithms, we will  
investigate end-to-end workflows based on the LDBC Graphalytics benchmark~\cite{DBLP:journals/pvldb/IosupHNHPMCCSAT16}.

Algorithmically we see a number of research directions to pursue.   With end-to-end workflows, the performance
of data ingestion heavily impacts performance.  We are interested in improving data ingestion performance
by exploiting a CPU's SIMD instructions~\cite{DBLP:journals/vldb/LangdaleL19}.  We are also interested in how  
LAGraph maps onto GPUs using versions of the GraphBLAS optimized for GPUs.