=Paper=
{{Paper
|id=Vol-1383/paper30
|storemode=property
|title=MapGraph - Graphprocessing at 30 Billion Edges per Second on NVIDIA GPUs
|pdfUrl=https://ceur-ws.org/Vol-1383/paper30.pdf
|volume=Vol-1383
|dblpUrl=https://dblp.org/rec/conf/semweb/FuDBBT14
}}
==MapGraph - Graphprocessing at 30 Billion Edges per Second on NVIDIA GPUs==
<pdf width="1500px">https://ceur-ws.org/Vol-1383/paper30.pdf</pdf>
<pre>
MapGraph - Graph processing at 30 billion edges per second on NVIDIA GPUs


Zhisong Fu             Harish Kumar Dasari    Bradley Bebee      Martin Berzins       Bryan Thompson
SYSTAP, LLC            University of Utah     SYSTAP, LLC        University of Utah   SYSTAP, LLC
fuzhisong@systap.com   hdasari@sci.utah.edu   beebs@systap.com   mb@sci.utah.edu      bryan@systap.com
                                                                                      (Presenting)

MapGraph is a disruptive technology that delivers extreme performance for graph problems on
many-core hardware. MapGraph can be run on a laptop, on EC2 HPC GPU compute nodes, and
on large GPU compute clusters. With processing speeds of up to 3 billion edges per second on a
single GPU, MapGraph changes what is possible with your data.

Many-core computing is the future. CPU architectures are not getting any faster. Continued
performance gains must come from many-core technologies such as GPUs or the Intel Xeon
Phi. GPUs are widely known for their role in games, high-performance computing, and high
FLOPS/watt ratio. However, graph algorithms are data-intensive, not compute intensive, and
have degree dependent parallelism. As a consequence, graph algorithms place an extreme
burden on the memory bus and communications network.

SYSTAP and the Scientific Computing and Imaging Institute have developed a capability for
extreme performance parallel graph algorithms on GPUs from laptops to large GPU
clusters. MapGraph provides a scalable technology for data-intensive workloads that addresses
the data-dependent parallelism, memory, and communication bottlenecks. I will review this
research and present our roadmap for this technology.

Learn more at http://mapgraph.io.

This work was (partially) funded by the DARPA XDATA program under AFRL Contract #FA8750-
13-C-0002. This material is based upon work supported by the Defense Advanced Research
Projects Agency (DARPA) under Contract No. D14PC00029. The authors would like to thank Dr.
White, NVIDIA, and the MVAPICH group at Ohio State University for their support of this work.

</pre>