Nathan R. Tallent

Nathan Tallent is a chief computer scientist in the Future Computing Technologies Group within Advanced Computing, Mathematics, and Data Division at Pacific Northwest National Laboratory. He joined PNNL in 2011.

General

Dr. Tallent is an internationally recognized expert in extreme performance. He understands all levels of performance, from massively scalable computing to chip pipelines; all system components, from interconnects, storage, memory, and processors; and workloads ranging from AI/ML, data analytics/graphs, and HPC.

Bio

Dr. Tallent's research is motivated by emerging challenges in distributed systems, scientific workflows, machine learning, and data management. He leads activities in continuum computing and the Performance Lab for EXtreme Computing and daTa where his contributions have spanned the challenges of performance measurement, modeling, bottleneck diagnosis, and optimization, with special attention to bottlenecks in networks, storage, and memory. He has made notable contributions to performance tools, both for performance modeling and for parallel performance analysis. He has more than 80 peer-reviewed publications, serves on several reviewing committees, and received a DOE Early Career award. He is one of the original developers of HPCToolkit, a widely used suite of performance tools on supercomputers. He received a Ph.D. in 2010 from Rice University.

Software

Dr. Tallent has led development of several research software prototypes for distributed AI systems, scientific workflows, and performance analysis and prediction.

• Distributed Workflows
  • DataFlowDrs is a new comprehensive suite of tools (DataLife, DaYu, FastFlow, FlowForecaster, QoSFlow) for performance optimization of scientific HPC workflows that automates several previously difficult manual analyses and substantially reduces the impact of data flow bottlenecks.
  • TAZeR, BigFlowSim: Remote I/O framework; and workflow I/O simulator-emulator and trace generator.
• AI Systems • Data Analytics
  • Rudder (over MassiveGNN): A framework for scaling graph neural networks to new levels with adaptive communication-efficient training for massive (distributed) GNNs within the state-of-the-art Amazon DistDGL (distributed Deep Graph Library).
  • PowerMorph and PowerTrip, for addressing the power constraints of large-scale training with federated heterogeneous datacenter power and intelligent adaptation of demand-response power.
  • High performance AI compression that preserves key features in scientific images (e.g., texture) and also achieves best-in-class image quality, compression ratio, and speed.
• Hardware/Software Co-design • Application Performance Analysis
  • MemGaze/MemFriend, a memory analysis toolset that combines low-overhead measurement; sophisticated, high-resolution trace analysis that reveals optimization opportunities by examining memory behavior from temporal, sequential, spatial, and relational viewpoints; and an architectural modeling framework, driven by sampled traces, that incorporates temporal interpolation and confidence-based filtering to evaluate cache and memory configurations.
  • OCEAN (Open-source CXL Emulation at Hyperscale Architecture and Networking), an emulation framework for CXL 3.0 that can evaluate fabric management, dynamic memory allocation, and coherent memory sharing across multiple hosts in advance of real hardware availability.
  • Palm: Palm is a suite of performance modeling tools (Palm, Palm-Task, Representative-Paths, Palm/FastFootprints, MIAMI-NW) to assist performance analysis and predictive model generation.
  • HPCToolkit: HPCToolkit is an integrated suite of tools for measurement and analysis of program performance on computers ranging from multicore desktop systems to GPU-accelerated supercomputers.
• Workload Benchmarking and Characterization
  • SEAK Suite: The SEAK Suite is a collection of constraining problems for common embedded computing challenges.
  • PERFECT Suite: The PERFECT Suite consists of kernels and applications for evaluating tradeoffs between performance, power, and architecture within the domains of radar and image processing.

He contributed to OpenAD (info), a tool for automatic differentiation (AD) of numerical computer programs.