High performance computing and data integrity

If a machine gives you the wrong answer and no error, how would you ever know?

The hardest fault in a large computing system is the one that produces a plausible wrong answer and raises nothing. At the scale modern simulation and analysis run, that is not a rare event. The center works on detecting corruption that leaves no trace, on encoding data so it stays usable and verifiable, and on the I/O paths that decide whether a result can be reproduced at all.

What the group builds

Silent data corruption detection

Corruption that produces a wrong answer without an error signal, detected from hardware performance counters at a cost low enough to leave running in production.

Reliable and efficient data encoding

Encoding schemes that keep simulation and analysis data usable, compact, and verifiable at extreme scale, funded by an NSF CAREER award.

Data integrity for HPC datasets

Using the sparsity structure of scientific datasets to find and correct corruption, with an NSF OAC Core award held jointly with the hardware security thrust.

Parallel I/O for data-intensive science

The storage and I/O paths that decide whether a large computation can be checkpointed, restarted, and reproduced.

Other thrusts

Most projects cut across two or three of them.

Smart grid cybersecurity and resilience AI and agentic systems for cyber-physical control Next-generation optical and 6G transport Fault-tolerant distributed and edge computing Hardware security and trusted devices Connected transportation, health, and infrastructure Nuclear energy and security