Future I/O systems for increasingly data-intensive computing environments face a challenging set of requirements. Data extraction
must be efficient, fast, and flexible; on-demand data annotation --
metadata creation and management -- must be possible without modifying
application code; and data products must be available for concurrent
use by multiple downstream applications (such as visualization and
storage), requiring consistency management and scheduling. In this
talk, I will present a collection of techniques designed to address
these challenges by decoupling data operations in space and in time
from core application codes. Our research results show that these
techniques can extract data efficiently and without perturbing compute
operations, that they can be used to perform application-specific
transformations while maintaining acceptable I/O bandwidth and
avoiding back-pressure, and that they can decouple "in-band" and
"out-of-band" processing to improve overall I/O performance.
Patrick M. Widener is a Senior Research Scientist in the Center for Comprehensive Informatics and Research Assistant Professor in the Department of Biomedical Engineering at Emory University. His research interests include experimental systems, I/O and storage software for large-data environments, middleware, and the generation and use of metadata. Dr. Widener received his Ph.D. in Computer Science from the Georgia Institute of Technology in 2005, and prior to beginning his Ph.D. studies he was employed as a software developer by several companies which no longer exist. He also holds a Master of Computer Science degree from the University of Virginia (1992), and a Bachelor of Science in Computer Science from James Madison University (1990).