No abstract

Unstructured meshes are used in many engineering applications with irregular domains, from elastic deformation problems to crack propagation to fluid flow. Because of their complexity and dynamic behavior, the development of scalable parallel software for these applications is challenging. The Charm++ Parallel Framework for Unstructured Meshes allows one to write parallel programs that operate on unstructured meshes with only minimal knowledge of parallel computing, while making it possible to achieve excellent scalability even for complex applications. Charm++'s messagedriven model enables computation/communication overlap, while its run-time load balancing capabilities make it possible to react to the changes in computational load that occur in dynamic physics applications. The framework is highly flexible and has been enhanced with numerous capabilities for the manipulation of unstructured meshes, such as parallel mesh adaptivity and collision detection. 1

Abstract-We present and analyze two new communication libraries, cudaMPI and glMPI, that provide an MPI-like message passing interface to communicate data stored on the graphics cards of a distributed-memory parallel computer. These libraries can help applications that perform general purpose computations on these networked GPU clusters. We explore how to efficiently support both point-to-point and collective communication for either contiguous or noncontiguous data on modern graphics cards. Our software design is informed by a detailed analysis of the actual performance of modern graphics hardware, for which we develop and test a simple but useful performance model.

No abstract

No abstract

SUMMARYWe present efficient support for generalized arrays of parallel data driven objects. Array elements are regular C++ objects, and are scattered across the parallel machine. An individual element is addressed by its 'index', which can be an arbitrary object rather than a simple integer. For example, an array index can be a series of numbers, supporting multidimensional sparse arrays; a bit vector, supporting collections of quadtree nodes; or a string. Methods can be invoked on any individual array element from any processor, and the elements can participate in reductions and broadcasts. Individual elements can be created or deleted dynamically at any time. Most importantly, the elements can migrate from processor to processor at any time. We discuss support for message delivery and collective operations in the face of such dynamic behavior. The migration capabilities of array elements have proven extremely useful, for example, in implementing flexible load balancing strategies and for exploiting workstation clusters adaptively. We present the design, an implementation, and performance results.

No abstract

We present efficient support for generalized arrays of parallel data driven objects. Array elements are regular C++ objects, and are scattered across the parallel machine. An individual element is addressed by its 'index', which can be an arbitrary object rather than a simple integer. For example, an array index can be a series of numbers, supporting multidimensional sparse arrays; a bit vector, supporting collections of quadtree nodes; or a string. Methods can be invoked on any individual array element from any processor, and the elements can participate in reductions and broadcasts. Individual elements can be created or deleted dynamically at any time. Most importantly, the elements can migrate from processor to processor at any time. We discuss support for message delivery and collective operations in the face of such dynamic behavior. The migration capabilities of array elements have proven extremely useful, for example, in implementing flexible load balancing strategies and for exploiting workstation clusters adaptively. We present the design, an implementation, and performance results. 372O. S. LAWLOR AND L. V. KALÉ is slow and duplicates delivery effort), misdelivered (delivered to the current occupant of that address), or lost (returned to the sender or dropped completely).MPI also uses physical addressing-messages in MPI are sent to a particular process number and tag. Like the postal service, if the computational entity the message is destined for moves, the message will either have to be manually forwarded or (more likely) will be misdelivered or lost. This means computational entities in MPI programs either never move, or can only be moved with a great deal of intricate, error-prone programming effort.While it is easy to deliver messages to a physical address, it is evident that moving objects are difficult to support. We present a simple, well-known solution to this problem-a layer of indirection between objects and physical addresses.Our framework allows an object to be referenced by a globally unique, problem-domain, userassigned logical address called an 'array index'. All communication is via the array index, which allows the object to be migrated in a completely general and user-transparent way.In this paper, we show how to support message delivery, creation, deletion, and migration scalably and efficiently using this logical addressing scheme. We present a solution to the problem of broadcasts and reductions in the presence of ongoing migrations. Finally, we present performance data from several actual applications built using this system. Partition decompositionMany of today's emerging high-end parallel applications are characterized by irregular and dynamic computational structure. These problems demand techniques such as latency hiding and dynamic load balancing to achieve good parallel performance. However, it requires significant programming effort to incorporate these techniques into a parallel program written using the prevalent processor-centric programming model exemplified by MPI.We a...