Ben completed his Ph.D. in 2020 at the University of Wisconsin-Madison where he studied Computer Science and was advised by Barton Miller. The focus of his research was on GPU performance tools (Diogenes) and in the development of extreme-scale density based clustering algorithms (Mr. Scan). Ben has spent time working (in an intern, adjunct, or full time capacity) at Lawrence Livermore National Laboratory, Argonne National Laboratory, and NASA's Goddard Space Flight Center. In May 2020, Ben will be joining Facebook as a Research Scientist.
Researched new performance tool techniques that led to the development of the performance tool Diogenes. Diogenes is a performance tool that targets GPU applications and uses a new feed-forward measurement (FFM) approach to data collection and analysis. FFM is a multi-run/multi-stage binary instrumentation approach where the applications execution behavior influences what instrumentation is inserted (and the data that is collected) on subsequent runs of the program. FFMs application guided approach to instrumentation allows Diogenes to detect previously hidden (from both other tools and developers) performance problems in GPU applications.
Developed a new extreme scale clustering method performing the Density-based spatial clustering of applications with noise (DBSCAN) algorithm. DBSCAN groups points in a dataset together based on locality (distance) from one another and a density characteristic (how many points fall within that distance). When the density characteristic is reached, the points are considered members of the same cluster. Our implementation of DBSCAN, called Mr. Scan, was the first DBSCAN-based algorithm to scale to 8192 GPU-equipped nodes and the first to process billions of points accurately. All other known implementations were limited to less than 100 nodes and had a maximum point count of 100 million points. Since the publication of this work, other researchers have taken the techniques we have developed and applied them to single node DBSCAN computations, resulting in an improvement in execution speed from 600K points in a hour to millions of points in a few minutes.
Porting the binary instrumentation components of Diogenes to PowerPC architectures. This work required the addition of support for PowerPC 8/9 binary rewriting (both static and dynamic) and stackwalking out of instrumented frames to the binary instrumentation tool Dyninst.
Development and research work on the I/O forwarding scalability layer (IOFSL). IOFSL was a tool that used a tree-based overlay network to scalable write to (and read from) a distributed file system. My research contribution to this project was on the use data compression to increase throughput to/from the filesystem.
Development of an experimental bittorrent server to serve geospatial data to collaborators at different institutions.
Development of new techniques for geospatial data analysis culminating in a web-based interface for processing satellite imagery data on demand, overlaying this data onto various web-based mapping tools (Google Maps and ArcGIS).
Benjamin Welton and Barton P. Miller, "Identifying and (Automatically) Remedying Performance Problems in CPU/GPU Applications", The 2020 International Conference on Supercomputing (ICS), Barcelona Spain, June 2020 [PDF]
Benjamin Welton "The Feed-Forward Measurement Model: A Performance Measurement Model for CPU/GPU Architectures", PhD Dissertation, Madison WI, March 2020 [PDF]
Benjamin Welton and Barton P. Miller, "Diogenes: Looking For An Honest CPU/GPU Performance Measurement Tool", Supercomputing 2019 (SC 2019), Denver CO, November 2019 [PDF]
Benjamin Welton and Barton P. Miller, "Exposing Hidden Performance Opportunities in High Performance GPU Applications", 18th IEEE/ACM International Symposium on Cluster, Cloud and Grid Computing (CCGrid), Washington DC, May 2018. (Best Paper) [PDF]
Benjamin Welton and Barton P. Miller, "Data Reduction and Partitioning in an Extreme Scale GPU-Based Clustering Algorithm", 2nd International Workshop on Data Reductions for Big Scientific Data (DRBSD), Denver CO, November 2017. [PDF]
William R. Williams, Xiaozhu Meng, Benjamin Welton, and Barton P. Miller, "Dyninst and MRNet: Foundational Infrastructure for Parallel Tools", 9th Annual Parallel Tools Workshop, Dresden Germany, September 2015. [PDF]
Benjamin Welton and Barton P. Miller, "The Anatomy of Mr. Scan: A Dissection of Performance of an Extreme Scale GPU-Based Clustering Algorithm", Workshop on Latest Advances in Scalable Algorithms for Large-Scale Systems (ScalA '14) New Orleans LA, November 2014. [PDF]
Benjamin Welton, Evan Samanas, and Barton P. Miller, "Mr. Scan: Extreme Scale Density-Based Clustering using a Tree-Based Network of GPGPU Nodes", Supercomputing 2013 (SC 2013), Denver CO, November 2013. [PDF]
Benjamin Welton, Dries Kimpe, Jason Cope, Christina M. Patrick,Kamil Iskra, and Robert Ross, "Improving I/O Forwarding Throughput with Data Compression", The 2011 IEEE International Conference on Cluster Computing (CLUSTER 2011), September 2011. [IEEE Open Access]
Benjamin Welton, Adam Milewski, Mohamed Sultan, and Kyle Chouinard, "Creation of a web-based GIS server with custom Geo-Processing tools for enhanced Hydrologic applications" (poster). American Geophysical Union, Fall Meeting 2010, abstract #H43B-1226
Mohamed Sultan, S. Metwally, Adam Milewski, Dee Becker, Mohamed Ahmed, William Sauck, Farouk Soliman, Neil Sturchio, Eugene Yan, Rashed Mohamed, Ahmad Wagdy, Richard Becker, and Benjamin Welton, "Modern recharge to fossil aquifers: Geochemical, geophysical, and modeling constraints", Journal of Hydrology, Volume 403, Issues 1-2, 6 June 2011, Pages 14-24, ISSN 0022-1694, DOI: 10.1016/j.jhydrol.2011.03.036.
Mohamed Ahmed, Mohamed Sultan, Jay Wahr, Eugene Yan, Adam Milewski, William Sauck. Richard Becker, and Benjamin Welton, "Integration of GRACE (Gravity Recovery and Climate Experiment) data with traditional data sets for a better understanding of the time- dependent water partitioning in African watersheds", Geology, Volume 39.5, 2011, Pages 479-82.
Mohamed Sultan, S. Metwally, Adam Milewski, Dee Becker, Mohamed Ahmed, William Sauck, Farouk Soliman, Neil Sturchio, Eugene Yan, Rashed Mohamed, Ahmad Wagdy, Richard Becker, and Benjamin Welton. Modern Recharge of the Nubian Aquifer: Remote Sensing, Geochemical, Geophysical, and Modeling Constraints. American Geophysical Union, Fall Meeting 2010, abstract #H21L-05
Peter Marsala, M. El Sayed, Mohamed Sultan, Jay Wahr, Adam Milewski, Richard Becker, Benjamin Welton, and Rajesh Balekai, "Integration of grace data with inferences from traditional datasets for a better understanding of the time-dependent water storage variability in large-scale aquifers: case studies from Africa". GSA annual meeting Portland 2009, Paper No. 227-1