[Colloquium] Laguna/MS Presentation/Oct 26, 2016

[Margaret Jaffey via Colloquium]

This is an announcement of Sean Laguna's MS Presentation.

------------------------------------------------------------------------------
Date:  Wednesday, October 26, 2016

Time:  3:15 PM

Place:  Ryerson 251

M.S. Candidate:  Sean Laguna

M.S. Paper Title: Using Persistent Data Models to Automate Parallelism
under Synchronization Requirements

Abstract:
We implement a vector, using a bit-partitioned trie, that allows for
consistent concurrent reading and writing between multiple tasks under
some constraints on the properties of those tasks. This data
structure, along with a work queue and a framework for resolving
operations that conflict with each other, yields a method for
automatically parallelizing a graph of tasks that may have
synchronization requirements under traditional memory models. We test
the method against other shared-memory parallelism techniques,
including C++ threads and OpenMP directives, by invoking repeated
reduce and foreach operations on an initial random input in the MB to
GB scale, as well as by implementing and benchmarking a solution to
the 1-dimensional heat equation. Our method achieves comparable
scaling results and similar raw performance. It beats serial
implementations on a dual-core machine and runs only 4% to 50% slower
than the best-performing parallel technique on up to 16 logical cores,
without requiring manual synchronization between steps and while
providing greater algorithmic flexibility.

Sean's advisor is Prof. L. Ridgway Scott

Login to the Computer Science Department website for details:
 https://www.cs.uchicago.edu/phd/ms_announcements#slaguna

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=
Margaret P. Jaffey            margaret at cs.uchicago.edu
Department of Computer Science
Student Support Rep (Ry 156)               (773) 702-6011
The University of Chicago      http://www.cs.uchicago.edu
=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=