C11949: Fast Data Reduction via Multi-Processor FrameworkNovelty:
The current invention is an improved framework for reducing the size of input data that utilizes multiple processors running simultaneously (in parallel).
Value Proposition:
The amount of data available is growing exponentially. There are steepening requirements to process this data. In parallel computing (multiple processors executing instructions simultaneously), a reduction algorithm (used to reduce the size of input data) often requires more time than the theoretic ideal. Furthermore, it suffers from the unbalance of workload in each parallel thread, which results in a waste of computing power. The current invention is a framework for parallel reduction that can exploit the wasted computing power and can finish a reduction faster than current methods. Advantages of the current invention include:
• Can be 3 times faster than current state-of-the-art parallel reduction approaches.
• Shows how to achieve the optimum performance in the reduction problem, and can fully solve the problem of unbalanced workload.
• Potentially 100 times faster computation than sequential methods in certain problems that involve reduction.
Technical Details:
Johns Hopkins researchers have developed a framework for the reduction of data which shows significant improvements over alternatives. The framework has been successfully implemented in several applications to show its potential. Also noteworthy is that the effectiveness level of the invention increases with the problem scale.
This framework can achieve the optimum performance for parallel reduction. Real-world implementation show the expected speed up as compared to the traditional reduction implementation.
Looking for Partners:
To develop and commercialize the technology as a framework for improving data reduction performance.
Stage of Development:
Prototype
Data Availability:
Under CDA/NDA
Publications/Associated Cases:
Not available at this time
