Parallelizing Data-Centric Programs
Abstract
Increasingly, the Air Force relies on data-centric software for strategic applications. We have studied data-centric software applications on unique types of datasets -- graphs, spatial data and collections of images. As a result of our work on graphs, we have developed a new high-performance parallel graph processing framework called GRACE. For spatial data, we have conducted a comprehensive benchmarking study of existing join processing algorithms and made our benchmark available to the public to promote further improvements for this important class of algorithms. Second, we have examined the applicability of general purpose cloud infrastructure for data-centric applications. We have conducted extensive performance studies and developed a jitter-tolerant runtime for tick-based applications as well as a deployment advisor for such applications. Third, we have studied the problem of parallel agents who need to communicate and coordinate to achieve a common goal. We have developed a novel abstraction called entangled queries that allows simple and efficient coordination in a wide variety of realistic scenarios.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 25, 2013
- Accession Number
- ADA590231
Entities
People
- Johannes E. Gehrke
Organizations
- Cornell University