Authors: Nathan Thomas

Venue: Doctoral dissertation, Parasol Laboratory, Department of Computer Science, Texas A&M University
DOI:
Link to Publication

Abstract:
Parallel programming is becoming mainstream due to the increased availability of multiprocessor and multicore architectures and the need to solve larger and more complex problems. Languages and tools available for the development of parallel applications are often difficult to learn and use. The Standard Template Adaptive Parallel Library (STAPL) is being developed to help programmers address these difficulties. STAPL is a parallel C++ library with functionality similar to STL, the ISO adopted C++ Standard Template Library. STAPL provides a collection of parallel pContainers for data storage and pViews that provide uniform data access operations by abstracting away the details of the pContainer data distribution. Generic pAlgorithms are written in terms of PARAGRAPHs, high level task graphs expressed as a composition of common parallel patterns. These task graphs define a set of operations on pViews as well as any ordering (i.e., dependences) on these operations that must be enforced by STAPL for a valid execution. The subject of this dissertation is the PARAGRAPH Executor, a framework that manages the runtime instantiation and execution of STAPL PARAGRAPHS. We address several challenges present when using a task graph program representation and discuss a novel approach to dependence specification which allows task graph creation and execution to proceed concurrently. This overlapping increases scalability and reduces the resources required by the PARAGRAPH Executor. We also describe the interface for task specification as well as optimizations that address issues such as data locality. We evaluate the performance of the PARAGRAPH Executor on several parallel machines including massively parallel Cray XT4 and Cray XE6 systems and an IBM Power5 cluster. Using tests including generic parallel algorithms, kernels from the NAS NPB suite, and a nuclear particle transport application written in STAPL, we demonstrate that the PARAGRAPH Executor enables STAPL to exhibit good scalability on more than $10^4$ processors. 

@phdthesis{Thomas-tpdaio-2012, 
 address = {Stanford, CA}, 
 author = {Thomas, Nathan}, 
 month = {January}, 
 school = {College Station, TX, USA}, 
 title = {Texas A&M University}, 
 year = {2012} 
}