We present a tool that allows to run an MPI application on several MPPs without having to change the application code. PACX (PArallel Computer eXtension) provides to the user a distributed MPI environment with most of the important functionality of standard MPI. It is therefore well suited for usage in metacomputing. We are going to show how two MPPs are configured by PACX into a single virtual machine. The underlying communication management that makes use of highly optimized MPI for internal communication and uses standard protocols for external communication is presented. The performance of PACX for several basic message-passing calls is described. This covers latency, bandwidth, synchronization and global communication.

Interoperable MPI (IMPI) is a protocol specification to allow multiple MPI implementations to cooperate on a single MPI job. Unlike portable MPI implementations, an IMPI-connected parallel job allows the use of vendor-tuned message passing libraries on given target architectures, thus potentially allowing higher levels of performance than previously possible. Additionally, the IMPI protocol uses a low number of connections, which may be suitable for parallel computations across WAN distances. The IMPI specification defines a low-level wireline protocol that MPI implementations use to communicate with each other; each point-to-point and collective function in MPI-1 automatically uses this low-level protocol when communicating with remote a MPI implementation. When running IMPI jobs, the only change visible to the user is the sequence of steps necessary to run the job; any correct MPI program will run correctly under IMPI. In this paper, we provide an overview of IMPI, describe its incor...

Abstract not found

MPI is the de facto message passing standard for multicomputers and networks of workstations, established by the MPI Forum, a group of universities, research centers, and national laboratories (from both the United States and Europe), as well as multinational vendors in the area of high performance computing. MPI has been implemented already by several groups. Worldwide acceptance of MPI has been quite rapid. This paper overviews several areas in which MPI can be extended, discusses the merits of making such extensions, and begins to demonstrate how some of these extensions can be made. In some areas, such as intercommunicator extensions, significant progress has been made by us already. In other areas (such as remote memory access), we are merely proposing extensions to MPI that we have not yet reduced to practice. Furthermore, we point out that other researchers are evidently working in parallel with us on their own extension concepts for MPI. 1 Introduction The MPI Forum introduced...

This paper presents a new kind of portability system, Unify, which modifies the PVM message passing system to provide (currrently a subset of) the Message Passing Interface (MPI) standard notation for message passing. Unify is designed to reduce the effort of learning MPI while providing a sensible means to make use of MPI libraries and MPI calls while applications continue to run in the PVM environment. We are convinced that this strategy will reduce the costs of porting completely to MPI, while providing a gradual environment within which to evolve. Furthermore, it will permit immediate use of MPI-based parallel libraries in applications, even those that use PVM for user code. We describe several paradigms for supporting MPI and PVM message passing notations in a single environment, and note related work on MPI and PVM implementations. We show the design options that existed within our chosen paradigm (which is an MPI interface added to the base PVM system), and why we chose that par...

Users of parallel machines need to have a good grasp for how different communication patterns and styles affect the performance of message-passing applications. LogGP is a simple performance model that reflects the most important parameters required to estimate the communication performance of parallel computers. The message passing interface (MPI) standard provides new opportunities for developing high performance parallel and distributed applications. In this paper, we use LogGP as a conceptual framework for evaluating the performance of MPI communications on three platforms: Cray-Research T3D, Convex Exemplar 1600SP, and a network of workstations (NOW). Our objective is to identify a performance model suitable for MPI performance characterization and to compare the performance of MPI communications on several platforms.

This paper describes our experience in developing and maintaining multi-platform parallel / distributed scientific computing software. We gained this experience while developing the parallel computation environment for //ELLPACK

Device Interface (ADI), which provides a small yet efficient interface to specific hardware and software of a particular manufacturer. Computer vendors need only to implement particular interface on their system and still get the benefits of peak efficiency. As a software, MPICH promotes the adoption of the MPI Standard by providing a starting point for vendor proprietary implementations. The portabilityof MPICH means that it can be used on all of current parallel environments, including parallel computers and clusters of workstations. It narrows the performance gap between user level (MPI) programs and the capabilities of the hardware. MPICH currently runs both in a native mode and over TCP/IP sockets: ffl IBM SPx ffl Intel Paragon ffl Cray T3D, Cray YMP, Cray PVP, Cray C90 ffl Meiko CS2 ffl TMC CM5 ffl NCUBE NCUBE-2 ffl Convex Exemplar ffl SGI (Power) Challenge ffl Sun Multiprocessors ffl 486's running Linux or FreeBSD ffl Sun, HP, SGI, IBM (RS/6000) and DEC (Alpha) works...

We propose a constraint-based algorithm for Bayesian network structure learning called recursive autonomy identification (RAI). The RAI algorithm learns the structure by recursive application of conditional independence (CI) tests of increasing orders, edge direction and structure decomposition into autonomous substructures. In comparison to other constraintbased algorithms d-separating structures and then directing the resulted undirected graph, the RAI algorithm combines the two processes from the outset and along the procedure. Learning using the RAI algorithm renders smaller condition sets thus requires a smaller number of high order CI tests. This reduces complexity and run-time as well as increases accuracy since diminishing the curse-of-dimensionality. When evaluated on synthetic and "real-world " databases as well as the ALARM network, the RAI algorithm shows better structural correctness, run-time reduction along with accuracy improvement compared to popular constraint-based structure learning algorithms. Accuracy improvement is also demonstrated when compared to a common search-and-score structure learning algorithm. 1

To include interactions with human users in simulations of the use of products, the most common approach is to couple human subjects to the behavioral product model in the simulation loop using interfaces based on VR and haptics. Replacing human subjects by human models with simulation capabilities could offer a cost-saving alternative. Currently available human models have not yet been deployed this way. This paper explores the possibilities to achieve mutual closed-loop coupling between human models and artifact models for enabling fully software-based interaction simulations. We have not only investigated human control in simulations, but also solutions to include embedded control in artifacts. The paper critically reviews existing (partial) solutions to simulate or execute control behaviors, and to close the control loops we identified in human-artifact interaction simulation. We concluded that closed-loop control of interaction simulations can be achieved by selectively combining existing partial solutions. Inclusion of decision-making appears to be the biggest challenge. Promising solutions are (i) cognitive simulation and (ii) execution of conjectured interactions specified as logical instructions, typically in the form of scenarios. Based on scenarios, which we expect to be more intuitive for designers, a new approach is now being developed.