![]() |
Home | People | Research | Publications | Tech Transition | Events | Partnerships | About ISR | Contact Us
![]() |
||
Exploiting Event Patterns to Reduce Delivery Costs in Scatter-Gather Parallel ProcessesStudent: David Woollard, USC/ISR
Advisor: Nenad Medvidović, USC/ISR and Theodore Berger, USC Department of Biomedical Engineering
Abstract: The scatter-gather event pattern is commonly used in many high-performance codes which rely on multipart data-parallel computations. There are a number of collective calls and algorithms available to developers using the ubiquitous Message Passing Interface (MPI) which perform various aspects of scattering and gathering. Unfortunately, reliance on these library calls often introduces excessive synchronization into codes, reducing overall performance. We will present ADaPT, an Adaptive Data-Parallel Publication/Subscription Transport Protocol, which reduces time lost to unnecessary synchronization associated with unicast event-passing commands employed to scatter these datasets. David Woollard holds a B.S. in Computer Engineering and Computer Science from the University of Southern California's Viterbi School of Engineering. He has received numerous awards for his academic research, including an Intel Research Grant and the Dean's Award for Outstanding Achievement. He is a Research Assistant in the Laboratory for Neural Dynamics at USC. He is a member of the Software Architectures Group within the Center for Software Engineering and is currently a graduate student in USC's Computer Science Department. |