Satish Puri
Missouri University of Science and Technology, MO·Computer Science·Early-career
Science now produces huge datasets, and many of them describe shapes. Examples include outlines of tissue in pathology slides, solar events on the Sun, geographic regions and movement paths.
- Fields
Computer Science
Engineering
- Topics
high performance computing
parallel computing
geospatial analytics
- Grants
- 4 known grants, through Aug 2027
What they work on
- Find which computing tasks and design settings gain the most speed on SmartSSD computational storage devices, which pair a drive with an FPGA chip, through a design space exploration on multicore systems.
- Build a programming framework with map, reduce, filter, join, scatter, gather and all-to-all operations for computational storage, and use it to run range queries, similarity searches and joins on points and polygons.
- Design processing-in-memory-aware versions of R-trees, KD-trees, locality sensitive hashing and product quantization, and build K-nearest-neighbor searches on UPMEM PIM hardware.
- Model the performance of PIM search algorithms with Roofline and iso-efficiency analysis, and develop a MapReduce-like interface for search-and-refine workloads.
- Develop approximate nearest-neighbor search for 10 to 100 billion polygons and trajectories, using product quantization and new LSH functions designed for Jaccard distance, which measures shape overlap.
- Create distributed-memory filter-and-refine algorithms for geospatial analysis that cut communication, using network-layout-aware data splitting and RDMA on data processing units, released within MPI-GIS.
More about the research
Finding similar shapes, or every shape inside a region, takes heavy computation. Speed is often limited by moving data between memory, storage and the processor rather than by the arithmetic itself. This research designs algorithms and software that perform these spatial searches faster on large scales. It works with new kinds of computer hardware that compute closer to the data. Across four projects, the shared goal is fast, scalable similarity and range searches on points, polygons and trajectories.
Why it matters
Faster and more scalable spatial search could help scientists who analyze huge shape datasets. Examples include pathologists comparing tumor tissue outlines, solar physicists predicting solar flares, and geoscientists or intelligence analysts locating shapes on maps. Open-source libraries and programming frameworks could make emerging hardware, such as SmartSSDs, processing-in-memory chips and data processing units, easier for researchers in these fields to use.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Collaborative Research: OAC: Approximate Nearest Neighbor Similarity Search for Large Polygonal and Trajectory Datasets
NSF · Aug 2023 – Aug 2027
CAREER: Communication-efficient and topology-aware designs for geo-spatial analytics on heterogeneous platforms
NSF · Jul 2023 – Jul 2027
Collaborative Research: OAC Core: Exploiting SmartSSD-based Computational Storage Architectures for Large Scale Similarity and Range Searches
NSF · Jul 2025 – Jun 2027
Closest labs
Cristinel Ababei
Marquette University, WI·Computer Science
Explore the design space of systems pairing multi-core CPUs with SmartSSD computational storage devices to identify the computational kernels and parameters that give the best acceleration.
Michael Bender
Stony Brook University, NY·Computer Science
Design efficient methods to build graph-based vector search indexes such as HNSW and DiskANN, reducing their currently very high construction costs.
Mahmut Kandemir
Pennsylvania State University, PA·Computer Science
Characterize diverse HPC workloads on real and simulated computational storage devices to identify which computations benefit from running inside storage, and what performance tradeoffs result.
The school
Missouri University of Science and Technology: 7 labs on PI Pages, $70M a year on research, 96 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.