Mahmut Kandemir
Pennsylvania State University, PA·Computer Science
Supercomputers increasingly spend time and energy just shuttling huge datasets between storage and processors. As simulations merge with machine learning and data analytics, this data movement has become a major bottleneck.
- Fields
Computer Science
Engineering
- Topics
computational storage
near-storage computing
high-performance computing
- Grants
- 1 known grant, through Jun 2028
What they work on
- 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.
- Design software abstractions and interfaces that expose each CSD's capabilities, including its compute power, memory capacity and differences from other devices, to compilers and applications.
- Develop compiler-directed optimizations for placing code and data, parallelizing loops and migrating data between CSDs, coordinated with a runtime system that schedules offloaded code across multiple applications.
- Build a simulation platform for future CSD architectures and release a benchmark suite for evaluating CSD-aware optimizations.
More about the research
Computational storage devices (CSDs) offer a fix: storage hardware, such as solid-state drives, with built-in processors that can compute right next to the data. But CSDs are hard to program, because current tools are low-level, hardware-specific and lack standard abstractions. This project builds software that would make CSDs practical for high-performance computing (HPC).
Why it matters
Faster, more energy-efficient processing of large datasets could speed data-heavy scientific work in fields the project names, including energy systems, AI, astrophysics and precision health. Making computational storage easier to program could help supercomputing centers and hardware makers actually use the technology. Open-source compilers, runtimes and benchmarks would give other researchers tools to build on.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Feng Chen
Indiana University, IN·Computer Science
Develop hardware functionality and system-level memory management for in-memory computing over Compute Express Link (CXL), the interconnect standard that lets processors and memory devices share memory, to reduce fragmentation of memory resources.
Peng Jiang
University of Iowa, IA·Computer Science·Early-career
Develop a runtime system that gives a virtualized view of data partitioned across heterogeneous devices, plus a compiler tool that places data using a statistical performance model of random memory accesses.
Xian-He Sun
Illinois Institute of Technology, IL·Computer Science
Engineer IOWarp, a modular and scalable data management platform for AI-infused scientific workflows, using NVMe SSDs, CXL memory devices and CPU-GPU co-design to reduce data access times on HPC systems.
The school
Pennsylvania State University: 60 labs on PI Pages, $1.3B a year on research, 743 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.