Dr. Xian-He Sun is the Founding Director of the Gnosis Research Center, an IEEE Fellow (2012), and a University Distinguished Professor (2014) and the Ron Hochsprung Endowed Chair (2021) of Computer Science at Illinois Institute of Technology. Named to the ICPP Wall of Fame as one of the Top 50 contributors in the conference's first 50 years (2023) and recognized in the BenchCouncil Top 100 Chips Achievements (2023), his contributions include the memory-bounded performance model (Sun-Ni Law, 1990), the Concurrent Average Memory Access Time (C-AMAT) framework, published in IEEE Computer in 2014, and the Dataflow under the von Neumann Machine (dataflowV) architecture. The memory-bounded model appears in textbooks as the Sun-Ni Law. Dr. Sun currently serves as Editor-in-Chief of IEEE Transactions on Parallel and Distributed Systems. His archived SCS biography records a B.S. in Mathematics from Beijing Normal University, followed by an M.S. in Mathematics, an M.S. in Computer Science, and a Ph.D. in Computer Science from Michigan State University; it does not give degree years. Before Illinois Tech, he held a postdoctoral appointment at Ames National Laboratory, worked as a staff scientist at ICASE at NASA Langley Research Center, was an ASEE fellow at the U.S. Navy Research Laboratories, and served as an associate professor and founding director of the SCS Laboratory at Louisiana State University in Baton Rouge. He joined Illinois Tech in August 1999. The same biography records his guest-faculty appointment in Argonne National Laboratory's Mathematics and Computer Science Division from 1999 to 2020 and his service as chair of Illinois Tech's Department of Computer Science from fall 2009 to summer 2014.
Research Vision
Computing has entered a data-centric era where the bottleneck is no longer computing but data processing. Dr. Sun's research program addresses this fundamental shift across the full stack, from memory system fundamentals (the Sun-Ni Law and C-AMAT model) through software system development (Hermes, LABIOS, ChronoLog) to emerging paradigms (Dataflow under von Neumann). His work establishes the mathematical foundations for memory-bounded data-centric thinking while simultaneously building the systems that translate theory into practice at extreme scale.
Top Collaborators
Featured Publications
Another View on Parallel Speedup
Presented at SC'90, this paper introduced the concept of memory-bounded speedup for parallel systems. Co-authored with Lionel Ni, it laid the groundwork for the Sun-Ni Law and challenged prevailing assumptions about parallel performance scaling.
Toward A Better Parallel Performance Metric
Co-authored with John Gustafson, this work proposed improved parallel performance metrics that account for problem scaling. It has been cited in HPC benchmarking work since.
Efficient Tridiagonal Solvers on Multicomputers
Developed efficient parallel algorithms for tridiagonal systems on multicomputers. Published in IEEE Transactions on Computers, this work demonstrated scalable numerical methods for distributed-memory machines.
Scalable Problems and Memory-Bounded Speedup
Introduced the memory-bounded speedup model, known as the Sun-Ni Law. It unifies Amdahl's law and Gustafson's scaled speedup as special cases, establishing one of the foundational results in parallel computing theory.
Scalability of Parallel Algorithm-Machine Combinations
Formally defined scalability for algorithm-machine combinations and proposed a quantitative measurement method. Published in IEEE TPDS, this work set the theoretical foundation for evaluating parallel system scalability.
Performance Modeling and Prediction of Non-Dedicated Network Computing
Developed a mathematical model for predicting performance in non-dedicated network computing environments. Published in IEEE Transactions on Computers, it addressed scheduling and resource sharing on privately owned workstation networks.
A QoS Guided Scheduling Algorithm for the Computational Grid
Introduced a QoS-guided min-min heuristic for grid task scheduling. It is a reference in distributed resource management and grid computing.
Grid Harvest Service: A System for Long-Term, Application-Level Task Scheduling
Presented Grid Harvest Service (GHS), a long-term task scheduling system for grid computing. It combined performance measurement, prediction, and scheduling into a unified framework for large-scale distributed applications.
lognP and log3P: Accurate analytical models of point-to-point communication in distributed systems
Introduced LognP and Log3P, analytical models of point-to-point communication that account for memory and middleware effects. Published in IEEE Transactions on Computers, these models improved communication cost prediction for clusters of SMPs.
Parallel I/O Prefetching Using MPI File Caching and I/O Signatures
Introduced I/O signature detection and MPI file caching for parallel prefetching. SC'08 Best Paper Finalist that launched the research trajectory toward GRC's I/O systems program.
Reevaluating Amdahl's Law in the Multicore Era
Revisited Amdahl's Law for multicore processors, showing that memory bandwidth and data access patterns fundamentally alter scaling behavior. Published in JPDC.
Memory Server
US Patent 7,865,570. A memory server architecture that provides data access as a service through aggressive prediction of client memory requirements. This patent embodies the data-centric philosophy of offloading data management from client processors.
Concurrent Average Memory Access Time
Defined Concurrent Average Memory Access Time (C-AMAT), extending the classical AMAT metric to account for data access concurrency at both component and system levels. Published in IEEE Computer.
Hermes: A Heterogeneous-Aware Multi-Tiered Distributed I/O Buffering System
Hermes provides multi-tiered I/O buffering across heterogeneous storage hierarchies. Co-authored with Anthony Kougkas, it is the predecessor of IOWarp.
LABIOS: A Distributed Label-Based I/O System
Presented LABIOS, a label-based I/O system achieving up to 17x speedup via asynchronous I/O and storage bridging. Co-authored with Anthony Kougkas. Karsten Schwan Best Paper Award at HPDC 2019.
DLIO: A Data-Centric Benchmark for Scientific Deep Learning Applications
DLIO is a data-centric benchmark for deep learning I/O workloads. Co-authored with the GRC team; the MLPerf Storage benchmark (MLCommons) is built on it. Best Paper at CCGrid 2021.
Research Interests
Selected Recent Projects
PhD Students Supervised at IIT
Awards & Honors
Selected Keynotes
Funded Research
