This study presents a hybrid concurrent programming model combining the previously developed Fork-Join model (FJM) and Actor model (AM). With the advent of multi-core computers, there is a renewed interest in programming models that reduce the burden of reasoning about and writing efficient concurrent programs. The proposed hybrid model shows how the divide-and-conquer approach of the FJM and the no-shared mutable state and event-driven philosophy of the AM can be combined to solve certain classes of problems more efficiently and productively than either of the aforementioned models individually. The hybrid model adds actor creation and coordination to into the FJM, while also enabling parallelization within actors. This study uses the Habanero-Java and Scala programming languages as the base for the FJM and AM respectively, and provides an implementation of the hybrid model as an extension of the Scala language called Habanero-Scala. The hybrid model adds to the foundations of parallel programs, and to the tools available for the programmer to aid in productivity and performance while developing parallel software.

Thesis (M.S.)--West Virginia University, 2002. / Title from document title page. Document formatted into pages; contains x, 85 p. : ill. (some col.), maps (some col.). Includes abstract. Includes bibliographical references (p. 62-69).

Thesis (MS)--University of Montana, 2009. / Contents viewed on January 15, 2010. Title from author supplied metadata.

No description available.

Thesis (D. Min.)--Midwestern Baptist Theological Seminary, 2000. / Includes bibliographical references (leaves 159-165).

Thesis (MS)--Montana State University--Bozeman, 2009. / Typescript. Chairperson, Graduate Committee: Dennis Neuman. Includes bibliographical references (leaves 124-138).

Thesis (D. Min.)--Midwestern Baptist Theological Seminary, 2000. / Includes bibliographical references (leaves 159-165).

Thesis (M.S.)--West Virginia University, 2005. / Title from document title page. Document formatted into pages; contains xi, 155 : ill., maps (some col.). Includes abstract. Includes bibliographical references (p. 80-91).

Esta Dissertação de Mestrado propõe dois novos algoritmos para tornar mais eficiente o escalonamento on-line de tarefas com dependências estritas em agregados de computadores que usam como middleware para troca de mensagens alguma implementação da MPI (até a versão 2.1). Esses algoritmos foram projetados tendo-se em vista programas construídos no modelo de programação fork/join, onde a operação de fork é usada sobre uma chamada recursiva da função. São eles: 1. O algoritmo RatMD, implementado através de uma biblioteca de primitivas do tipo map-reduce, que funciona para qualquer implementação MPI, com qualquer versão da norma. Utilizado para minimizar o tempo de execução de uma computação paralela; e 2. O algoritmo RtMPD, implementado através de um sistema distribuído sobre daemons gerenciadores de processos criados dinamicamente com a implementação MPICH2 (que implementa a MPI-2). Utilizado para permitir execuções de instâncias maiores de programas paralelos dinâmicos. Ambos se baseiam em roubo de tarefas, que é a estratégia de balanceamento de carga mais difundida na literatura. Para ambos os algoritmos apresenta-se modelagem téorica de custos. Resultados experimentais obtidos ficam dentro dos limites teóricos calculados. RatMD provê uma redução no tempo de execução de até 80% em relação ao algoritmo usual (baseado em round-robin), com manutenção do speedup próximo ao linear e complexidade espacial idêntica à popular implementação com round-robin. RtMPD mantém, no mínimo, o mesmo desempenho que a implementação canônica do escalonamento em MPICH2, dobrando-se o limite físico de processos executados simultaneamente por cada nó. / This Master’s Dissertation proposes two new algorithms for improvement on on-line scheduling of dynamic-created tasks with strict dependencies on clusters of computers using MPI (up to version 2.1) as its middleware for message-passing communication. These algorithms were built targeting programs written on the fork-join model, where the fork operation is always called over an recursive function call. They are: 1. RatMD, implemented as a map-reduce library working for any MPI implementation, on whatever norm’s version. Used for performance gain; and 2. RtMPD, implemented as a distributed system over dynamic-generated processes manager daemons with MPICH2 implentation of MPI. Used for executing larger instances of dynamic parallel programs. Both algorithms are based on the (literature consolidated) work stealing technique and have formal guarantees on its execution time and load balancing. Experimental results are within theoretical bounds. RatMD shows an improvement on the performance up to 80% when paired with more usual algorithms (based on round-robin strategy). It also provides near-linear speedup and just about the same space-complexity on similar implementations. RtMPD keeps, at minimum, the very same performance of the canonical MPICH2 implementation, near doubling the physical limit of simultaneous program execution per cluster node.

No description available.