Message Passing Interface Forum. MPI: A Message Passing Interface. In Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....this information is then used to construct a 14 parallel program satisfying these dependencies. Most recent work into parallel imperative languages centres around languages with explicit parallel constructs, like occam [INM84] or extensions to existing languages such as the MPI library [Mes93] for C. In functional languages introducing parallelism is slightly easier because they do not allow side e ects. Purely implicit approaches typically use evaluation transformation to determine what parts may be done in parallel. These methods include exploiting parallelism in graph reduction, ....

Message Passing Interface Forum. MPI: A message passing interface. In Proc. Supercomputing '93, pages 878-883. IEEE Computer Society, 1993.

....to add communication costs which could largely depend on communication performance characteristics of a real machine into the model and convert PRAM costs to real costs. 2. 2 Message Passing Model The message passing programming models provided by communications libraries such as PVM [37] or MPI [59] have been a dominant model for scientific and commercial parallel applications for the last decade. In this section, we first outline the basic concepts in message passing computing, and then we outline the two theoretical models, the BSP model and the LogP model which can model the cost of ....

....gather is used to describe having one process collect individual values from a set of processes. Gather is normally used after some computation has been done by these processes. Most message passing systems provide for these operations and other related operations. Message Passing Interface (MPI) [59] is an example of such communication routines. Processes communicate with one another by sending packets of information using the point to point communication routines such as MPI Send( and MPI Recv( MPI includes a wide selection of routines to offer different sorts of synchronisation in the ....

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proceedings of Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.

....All the computers had AMD K6 II processors at 450 MHz and 64 MB of RAM, running Redhat Linux version 6.2. They were connected by a private Fast Ethernet (100baseT) network. The software used to implement the parallel algorithms had to be well adapted to such a virtual parallel machine, and MPI [For93] or PVM [Sun90] would be the standard choices. Using either MPI or PVM is was impossible because most of their implementations are not suited for multithreaded use (not thread aware) The Athapascan 0 communication em multithreading runtime library [PGBP97] was chosen because: it was conceived ....

Message Passing Interface Forum. MPI: a message passing interface. Proceedings of the Supercomputing Conference, pages 878--885, November 1993.

....BSP libraries are shown. 1 Introduction Most message passing libraries are based on pairwise sends and receives. This means that for each send operation, a matching receive has to be issued on the destination processor. Widely available message passing libraries like PVM[GBD 94] and MPI[Mes93] provide the user with this kind of programming model. This approach, however, is very error prone because deadlocks can be easily introduced if a message is never accepted. Furthermore, it is dicult to determine the correctness and time complexity of programs implemented using pairwise sends and ....

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. Supercomputing '93, pages 878-883. IEEE Computer Society Press, 1993.

....object connected to the local sub domain simulator. This class uses member functions, given by each of the sub domain simulators, to retrieve and modify its local data. To handle the message passing we are using the C library Object Oriented MPI (OOMPI) 12] which is built on top of the MPI 1 [8]. This library has the great advantage that it can create communication objects and a processor topology which fit the geometry. We have made a manager class, named SOM Manager, for administrating each of the sub domain solvers. The class is controlled from the main program, and it has the global ....

Message Passing Interface Forum, MPI: A message Passing Interface. In Proc. of Supercomputing '93, pages 878-883. IEEE Computer Society Press, November 1993.

....SAMBA offers a set of functions that are used in many load balancing algorithms. Besides, the strategies in the load balancing library can be used as a basis for the development of new algorithms. A first version of the SAMBA framework has been implemented [PRR99] in C using the MPI library [MPI93]. Evaluation experiments have been executed on three different environments: an IBM SP2 system, a network of Sun workstations, and a cluster of IBM PCs. This paper also proposes classification criteria for loadbalancing algorithms designed for SPMD applications. The proposed set of criteria ....

Message Passing Interface Forum, "MPI: A message passing interface", Proceedings of Supercomputing'93 (A.Y. Zomaya, editor), 878--883, IEEE Computer Society, 1993.

....and address. It was quickly realised that message passing systems look much the same for any distributedmemory architecture, so it was natural to build standard interfaces to improve the portability of message passing programs. The most recent example of this is MPI (Message Passing Interface) [72, 153], which provides a rich set of messaging primitives, including point to point communication, broadcasting, and the ability to collect processes in groups and communicate only within each group. MPI has been defined to become the standard message passing interface for parallel applications and ....

Message Passing Interface Forum. MPI: A message passing interface. In Proceedings of Supercomputing '93, pages 878--883. IEEE Computer Society, 1993.

....the model eliminates the possibility of deadlock from parallel code; 2) one sided communication based upon remote memory access is used to exchange data among processors. This style of communication removes the intellectual burden of matching send receive pairs; 3) unlike either PVM [7] or MPI [11] the library is small with only eight subroutines; 4) it has a straightforward and accurate cost model. This approach to parallel programming is applicable to all kinds of parallel architecture: distributed memory architectures, shared memory multiprocessors, and networks of workstations. It ....

....have been performed by a single processor in the same time as to execute all the supersteps. The interpretation of g is not so simple, and is discussed in more detail in the next section. 3. 1 The computation to communication ratio g In traditional message passing systems such as PVM [7] and MPI [11] the cost of communication has to be considered in terms of sender receiver pairs. In contrast, the bulk aspect of the BSP model considers all the individual communications that occur within a superstep as a single monolithic unit. The cost of these communications is accurately modelled by ....

Message Passing Interface Forum. MPI: A message passing interface. In Proc. Supercomputing '93, pages 878--883. IEEE Computer Society, 1993.

....the execution of a specified function to deal with it. The network definition tools provide a graphic interface allowing the user the ability to cluster machines by architecture, control connectivity, and specify proportional computing capabilities. 4.1. 2 MPI MPI (Message Passing Interface) [76] is an effort by a group of vendors, computational scientists and computer scientists to consolidate the experiences gleaned from the use of message passing systems into a single standardized message passing system that will have the direct support of the computer vendors. Most of the message ....

....Ohio Supercomputer Center and provides a full implementation of the MPI standard for Fortran and C programs running on clusters of UNIX workstations. It also supports PVM compatibility, parallel I O, and debugging and monitoring tools. ffl MPI Model Implementation: The MPI Model Implementation [76, 53] was developed at Argonne National Laboratory to provide a full implementation of the MPI features, including both the Fortran and C bindings. ffl UNIFY: UNIFY [88] is a subset of MPI that has been built on top of PVM. UNIFY is a dual API (Application Program Interface) system. Users may write ....

[Article contains additional citation context not shown here]

Message Passing Interface Forum. MPI: A message passing interface. In Proc. Supercomputing '93, pages 878--883. IEEE Computer Society, 1993.

....under which a ULP must be blocked depends on the semantics defined by the message passing interface used by the application. For illustrative purposes, consider the MPI message passing interface, which defines a wide variety of blocking and non blocking communication primitives [MPIf93a, MPIf93b] Note that there are other message passing models such as active messages [vECGS92] that are neither equivalent nor subsumed by the MPI model. The intent of this section is give a flavour for the interaction between message passing semantics and scheduling. MPI defines blocking and non blocking ....

The Message-Passing Interface forum. MPI: A message-passing interface. In Supercomputing '93 Proceedings, pages 878--883, Portland, OR, November 1993.

....it shows that many iterations are required to solve these linear systems and that triangular solve takes at least 40 percent of runtime out of a total runtime which includes incomplete Cholesky and iterative solution. Iteration counts and timings are from BlockSolve running on top of MPI [Mes93] on a collection of Sun workstations connected by Ethernet. This system runs the same computation as would a full implementation of ICCG on Alewife. The data, partitions, and convergence would be the same. However, the overheads and latencies are substantially higher in the cluster than on ....

Message Passing Interface Forum. MPI: A message passing interface. In Supercomputing `93, pages 878--883. IEEE, 1993.

.... paradigm to Piranha or Paradise network models, which extend the basic Linda memory model to either adaptive parallelism or to fault tolerant shared memory servers, respectively [JUST96] The Message Passing Interface (MPI) The MPI specification was created by the Message Passing Interface Forum [MPIF93] with the goal of providing a portable parallel API that allows for efficient communication, heterogeneous implementations, convenient C and FORTRAN 77 language bindings, and consistency with current message passing paradigm practices (such as PVM, NX, p4, etc. Portability is achieved with the ....

Message Passing Interface Forum, "MPI: A Message Passing Interface," Proceedings of Supercomputing 1993, IEEE Computer Society Press, pp. 878-883, 1993.

....contain hidden barriers. Barrier synchronisation is also integral to the BSP programming model [12, 15, 19] in which computations proceed in supersteps which are separated by barriers. Barriers also appear in less structured parallel computation models such as PVM [3, 4, 9] pvm barrier) and MPI [8, 10, 17] (MPI barrier) Manufacturers have responded by including mechanisms for barrier synchronisation in today s parallel architectures, usually in software, but sometimes in hardware. One of the contributions of this paper is to show that such implementations should be treated with caution their ....

Message Passing Interface Forum. MPI: A message passing interface. In Proceedings of Supercomputing '93, pages 878--883. IEEE Computer Society, 1993.

....must be mapped to this abstract machine. This might be done by compilation but is often, in fact, done by implementing programming models as libraries. This rules out some potentially interesting designs, but simplifies implementation effort. An example of such a programming model is MPI [32]. From the perspective of the programmer, MPI provides a set of routines that can be called from a high level language. Commercial computer manufacturers have built implementations of MPI customised to the transport layer of their architectures. For example, MPI has been implemented for both ....

....We now demonstrate that these low level performance figures can be fully exploited at the application level. To do this we use the NAS Parallel Benchmarks 2. 1 [2] For the BSPlib benchmarks, we use codes ported by Oxford Parallel [30] In each case, we also include results from the MPI [32] versions of the code. On the IBM SP2 and the Origin 2000, these are proprietary MPI libraries, and on our cluster, mpich. A BSP implementation [30] of version 2.1 of the NAS parallel benchmarks was used to compare the performance of the BSPlib TCP, BSPlib UDP, and the BSPlib NIC implementations ....

Message Passing Interface Forum. MPI: A message passing interface. In Proc. Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.

....programs The BSP model encourages a disciplined use of computation and communication resources, in the sense that all processors perform lock step phases of computation followed by communication. One way of writing BSP programs is to use existing communication libraries such as PVM [3] or MPI [19] that support non blocking communications. These general purpose libraries, however, are rarely optimised for the relatively small, but by no means trivial, subset of operations that are required for representing the BSP programming paradigm [12, 21] To address this problem, the BSP research ....

Message Passing Interface Forum. MPI: A message passing interface. In Proc. Supercomputing '93, pages 878--883. IEEE Computer Society, 1993.

....Squyres y Brian C. McCandless y Andrew Lumsdaine y http: www.cse.nd. edu lsc research oompi 1 Introduction The emergence of the Message Passing Interface as a standard for parallel programming has been an important step forward in the development of parallel computing software technology [2]. At this time, however, language bindings for MPI have only been specified for C and Fortran 77. In this abstract, we provide a brief overview of the motivation, requirements, analysis, design, and implementation of Object Oriented MPI (OOMPI 1 ) a C class library for MPI. 2 Motivation ....

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. of Supercomputing '93, pages 878--883. IEEE Computer Society Press, November 1993.

....program description language. Thus, the OOMPI specification can also be considered as a generic object oriented class library specification which can thus also form the basis for MPI class libraries in other object oriented languages. 1. Introduction Although the Message Passing Interface (MPI) [3, 4, 9] is an increasingly popular and important standard for performance portable parallel programming, using MPI with C has been difficult because of the lack of standard C bindings and class libraries. There are three immediate alternatives for using MPI with C : use the existing C bindings, use ....

.... Port; OOMPI Message imsg(i, MY INT TAG) OOMPI Message jmsg(j, 10, MY TAG) Explicitly formed messages can be sent through the standard interface: Port.Send(imsg) Or they can be sent through the stream interface: Port jmsg; They can also be re used: i ; j[3] ; Port imsg jmsg; The OOMPI Datatype object is used to describe the datatype of a message. In addition to providing access to functions that build the less complicated userdefined datatypes such as MPI Type contiguous( and MPI Type vector( theOOMPI Datatype object also provides a ....

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. of Supercomputing '93, pages 878--883. IEEE Computer Society Press, November 1993.

....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 incorporation into the LAM implementation MPI, and show an example of its use. I. INTRODUCTION Since the publication of the MPI 1 standard [1], a large number of high quality MPI implementations have been made available. The ease of obtaining an MPI implementation has led to a new level of portability; parallel codes can now run on a variety of different operating systems and architectures simply by recompiling the same source code with ....

Message Passing Interface Forum, "MPI: A Message Passing Interface," in Proc. of Supercomputing '93, pp. 878--883, IEEE Computer Society Press, November 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. of Supercomputing '93, pages 878--883. IEEE Computer Society Press, November 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. Supercomputing, 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proceedings of Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Supercomputing '93, pages 878--883. IEEE Computer Society Press, 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. of Supercomputing '93, pages 878--883. IEEE Computer Society Press, November 1993.

No context found.

Message Passing Interface Forum. MPI: A Message Passing Interface. In Proc. of Supercomputing '93, pages 878--883. IEEE Computer Society Press, November 1993.

First 50 documents

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC