Dongarra, J. J., Hempel, R., Hey, A. J. G., and Walker, D. W. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report, TM-12231, Oak Ridge National Laboratory, Available at http://www.epm.ornl.gov/~walker/mpi1.ps, February 1993.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Such concerns are important during the design of the messaging protocol, which should remain simple as it is developed, evaluated, and refined. 23 2.3 Message Passing The basic concept behind message passing is that of processes communicating through messages. This paradigm is well understood [6, 10] and there are many implementations of this spanning various applications. For example, messages are used in a typical graphical windowing system (like X windows or Microsoft Windows) to convey events or commands between the windowing components and subsystems. Message passing can be e#ciently and ....

J.J. Dongorra, R. Hempel, A.J.G. Hey, and D.W. Walker. A proposal for a userlevel, message passing interface in a distributed memory model. Tech. Rep. TM12231, Oak Ridge National Laboratory, February 1993.

....Such concerns are important during the design of the messaging protocol, which should remain simple as it is developed, evaluated, and re ned. 23 2.3 Message Passing The basic concept behind message passing is that of processes communicating through messages. This paradigm is well understood [6, 10] and there are many implementations of this spanning various applications. For example, messages are used in a typical graphical windowing system (like X windows or Microsoft Windows) to convey events or commands between the windowing components and subsystems. Message passing can be eciently and ....

J.J. Dongorra, R. Hempel, A.J.G. Hey, and D.W. Walker. A proposal for a userlevel, message passing interface in a distributed memory model. Tech. Rep. TM12231, Oak Ridge National Laboratory, February 1993.

....to a standard message passing interface were discussed, and a working group established to continue the standardization process. A preliminary draft proposal, known as MPI1, was put forward by Dongarra, Hempel, Hey, and Walker in November 1992, and a revised version was completed in February 1993 [12]. MPI1 embodied the main features that were identified at the Williamsburg workshop as being necessary in a message passing standard. Since MPI1 was primarily intended to promote discussion and get the ball rolling, it focused mainly on point to point communications. MPI1 brought to the ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user- level, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....to a standard message passing interface were discussed, and a working group established to continue the standardization process. A preliminary draft proposal, known as MPI1, was put forward by Dongarra, Hempel, Hey, and Walker in November 1992, and a revised version was completed in February 1993 [12]. MPI1 embodied the main features that were identified at the Williamsburg workshop as being necessary in a message passing standard. Since MPI1 was primarily intended to promote discussion and get the ball rolling, it focused mainly on point to point communications. MPI1 brought to the ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a userlevel, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....The problem of how to achieve a reasonable load balancing in a practical and efficient way still remains an open problem. Although PVM plays the role of a de facto standard for the user and the software developer and even is evolving to an official message passing interface (MPI) standard [13], there is a lack of official standards or even agreed concepts at higher levels of abstraction for programming. We already explained above, that hypercomputers presently are unstable systems. Further efforts must be made to solve associated problems and to be able to cope with workstations, ....

J. J. Dongarra, R. Hempel, A. J. G. Hey and D. W. Walker, A Proposal for a User-Level, MessagePassing Interface in a Distributed Memory Environment. Technical report CS-93-186, University of Tennessee, Knoxville, january 1993.

.... At this point we note that our SPA and its associated decompositional solution seem to be especially suited for synchronous systems or systems with barrier synchronisation, such as they can be implemented with languages as Esterel [1] or with library based parallel programming systems such as MPI [7,20]. 3 Stochastic processes and all that In Section 3.1 we introduce discrete and continuous time Markov chains as well as semi Markov chains. We then introduce the new tagged semi Markov 7 chains and de ne a product in Section 3.2. Then, we comment on the usefulness of throughputs for the ....

Jack J. Dongarra, Rolf Hempel, Anthony J. G. Hey, and David W. Walker. A proposal for a user-level, message-passing interface in a distributed memory environment. Technical Report ORNL/TM-12231, Oak Ridge National Laboratory, Feb 1993.

....parallel and distributed systems support a message passing model in which processes operate asynchronously and communicate solely by posting matching send and receive commands. It is becoming increasingly common for these systems to support both blocking and nonblocking send and receive primitives [7, 8, 10, 16, 18, 19, 20], and some systems support both synchronous and asynchronous sends [7, 16] The synchronization and blocking properties of the send and receive primitives specify when they return to the calling process. A blocking asynchronous send does not return until the message has been copied out of the ....

....and communicate solely by posting matching send and receive commands. It is becoming increasingly common for these systems to support both blocking and nonblocking send and receive primitives [7, 8, 10, 16, 18, 19, 20] and some systems support both synchronous and asynchronous sends [7, 16]. The synchronization and blocking properties of the send and receive primitives specify when they return to the calling process. A blocking asynchronous send does not return until the message has been copied out of the sender s buffer, while a blocking synchronoussend does not return until the ....

[Article contains additional citation context not shown here]

J. Dongarra, R. Hempel, A. Hey, D. Walker, A Proposal for a User-Level, Message-Passing Interface in a Distributed Memory Environment, ORNL Technical Report, ORNL/TM-12231, June 1993.

....operation is usually defined as an individual event [1, 4, 10, 11, 14, 20, 23, 24] However, this approach may be too restrictive. Consider for example collective communication such as broadcast, reduce (all to one reduction) combine (all to all reduction) or concat (all to all broadcast) [2, 8, 12]. When modeling such kind of communication, it may be desirable to group several send and or receive operations in one event, since internal algorithms used to implement these operations are (or have to be) hidden to the user. Regarding the second issue, the semantic associated to the model is too ....

....always be received in the same order. Event e 2;1 is thus an elementary event. Figure 3: Program P decomposed in elementary events. As another example of elementary events, consider a collective communication operation like a combine 4 for example, where the result is known by all group members [8, 12]. A clean and correct implementation of such kind of operation is such that the final result only depends on the set of input data (one piece of data being specified by each process) i.e. on the set of states s i;j Gamma1 for each group member p i assuming the operation occurs in e i;j . In ....

J. Dongarra, R. Hempel, A. Hey, D. Walker, A Proposal for a User-Level, MessagePassing Interface in a Distributed Memory Environment, ORNL/TM-12231, March 1993.

....early 1990s. Probably the most widely used library based shared memory model is a set of macros for Fortran and C developed by researchers at Argonne National Laboratory [15] Widespread experience with library based message passing models has led to efforts to standardize on a single interface [26]. It is widely believed that shared memory programming models are easier to use than message passing models. This belief is supported by the dominance of (small scale) shared16 memory multiprocessors in the market, and by the many efforts by compiler writers [32, 68, 56] and operating system ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A Proposal for a UserLevel, Message-Passing Interface in a Distributed Memory Environment. ORNL/TM12231, October 1992.

.... The broadcasting of messages is a basic communication operation in message passing massively parallel processors (MPPs) Two variants of the broadcasting operation, the one to all and all to all broadcast, have been studied extensively and are generally included in communication libraries [1, 2, 6, 7, 12, 17]. In the one to all broadcast, one processor broadcasts a message to every other processor [4, 11, 15, 16, 18] In the all to all broadcast, every processor broadcasts a message to every other processor [3, 8, 10, 13, 20] Broadcasting problems arising in parallel applications are not limited to ....

....5 we discuss performance and scalability of the proposed algorithms on the Intel Paragon and Cray T3D. Section 6 concludes. 2 Algorithms without Repositioning A natural first approach for implementing communication operations is to make use of operations available in communication libraries [1, 2, 6, 7]. Not surprisingly, s to p broadcasting can be phrased in terms of such operations. We start this section by describing three such solutions. The broadcasting of the s messages can be performed by every source processor sending its message to a designated processor, say processor P 0 . P 0 ....

J.J. Dongarra, R. Hempe, A.J.G. Hey, D.W. Walker, A Proposal for a User-level, Message Passing Interface in a Distributed Memory Environment, Technical Report, TM 12231, Oak Rigde Laboratory, 1993.

....message passing; PVM (Parallel Virtual Machine) which supports message passing and was designed specifically for heterogeneous networks of workstations [33, 34] ffl 3.5 Types of System 3.5.1 Dedicated, Sequential System 3.5. 2 Non dedicated, Sequential System MPI (Message Passing Interface) [35, 36, 37, 38], which is supposed to bring together the main ideas from all of the other message passing systems, to create a single standard set of libraries for message passing in workstations clusters. With backing from many computer manufacturers, this system appears likely to become the standard method for ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....[19] At this workshop the basic features essential to a standard message passing interface were discussed, and a working group established to continue the standardization process. Following this a preliminary draft proposal, known as MPI1, was put forward by Dongarra, Hempel, Hey, and Walker [5]. This proposal was intended as a discussion document, and embodies the main features that were identified in the earlier workshop as being necessary in a message passing standard. A meeting of the mpi working group was held at Supercomputing 92, at which it was decided to place the ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....to a standard message passing interface were discussed, and a working group established to continue the standardization process. A preliminary draft proposal, known as MPI1, was put forward by Dongarra, Hempel, Hey, and Walker in November 1992, and a revised version was completed in February 1993 [12]. MPI1 embodied the main features that were identified at the Williamsburg workshop as being necessary in a message passing standard. Since MPI1 was primarily intended to promote discussion and get the ball rolling, it focused mainly on point to point communications. MPI1 brought to the ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a userlevel, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....[40] is becoming an industry wide standard for message passing. The MPI Forum consists of universities, research centers, national laboratories, and multi national high performance computing vendors. MPI was designed by over forty organizations and sixty people. The initial version, MPI1 [29], was published in June, 1993. The latest MPI (1.1) 55] was published in late 1994. MPI 2 will be published this year at Supercomputing 96. The goal is to create a universal communication language implemented on all parallel systems. The MPI Forum studied the existing message passing systems, and ....

Jack J. Dongarra, Rolf Hempel, Anthony J.G. Hey, and David W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report ORNL/TM-12231, Engineering, Physics, and Mathematics Division, Oak Ridge National Laboratory, Oak Ridge, TN, 1993.

....complicating the notion of receiving messages in the order in which they are sent . It is common for a single parallel computer or messagepassing library to support send and receive primitives with most, or even all, of the synchronization, blocking, and selectivity properties mentioned above [2, 5, 6, 7, 12, 13, 15]. As a result, it is essential to understand exactly how these properties interact with one another and with the message ordering properties. Unfortunately, the existing models of message passing systems are imprecise [2, 5, 6, 7, 13] or they model a very restricted set of send and receive ....

....blocking, and selectivity properties mentioned above [2, 5, 6, 7, 12, 13, 15] As a result, it is essential to understand exactly how these properties interact with one another and with the message ordering properties. Unfortunately, the existing models of message passing systems are imprecise [2, 5, 6, 7, 13], or they model a very restricted set of send and receive primitives [1, 3, 8, 9, 10, 11, 14] or they apply only to correct programs [3, 5, 11] While it might seem sufficient to consider only correct programs, there are actually two problems with this approach. First, large and complex ....

[Article contains additional citation context not shown here]

J. Dongarra, R. Hempel, A. Hey, D. Walker, A Proposal for a User-Level, Message-Passing Interface in a Distributed Memory Environment, ORNL/TM-12231, June 1993.

No context found.

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM12231, Oak Ridge National Laboratory, February 1993.

....standard, but it lacked collective communication routines, and was not thread safe in its management of resources. However, MPI 0 brought to the forefront a number of important issues and served as the initial catalyst for subsequent discussion. A revised version was published in February 1993 [8]. At the Minneapolis meeting, it was decided to place the standardization process on a more formal footing. The procedures and organization of the HPF Forum were adopted to create the MPI Forum. Subcommittees were formed for the major component areas of the standard and an email reflector was ....

J. Dongarra, R. Hempel, A. Hey and D. Walker, A proposal for a user-level, message passing interface in a distributed memory environment , Technical report ORNL/TM-12231, Oak Ridge National Laboratory (February 1993).

....to a standard message passing interface were discussed, and a working group established to continue the standardization process. A preliminary draft proposal, known as MPI1, was put forward by Dongarra, Hempel, Hey, and Walker in November 1992, and a revised version was completed in February 1993 [3]. MPI1 embodies the main features that were identified at the Williamsburg workshop as being necessary in a message passing standard. This proposal was intended to initiate discussion of standardization issues within the distributed memory concurrent computing community, and has served as a basis ....

....example. Finally, the migrating particles are sent to their destination process, dest, by a call to MPI send. The offsets in the Ztype datatype are interpreted relative to the address of the start of the particle array. struct Pstruct particle[1000] MPI datatype Ptype, Ztype; MPI datatype Stype[3]=fMPI double, MPI double, MPI intg; int Sblock[3] f1, 1, 1g; int Sindex[3] int Pindex[100] int Pblock[100] Sindex[0] 0; Sindex[1] sizeof(double) Sindex[2] 2 sizeof(double) MPI type struct (3, Stype, Sindex, Sblock, Ptype) j=0; for (i=0;i 1000;i ) if (x[i] right edge) f Pindex[j] ....

[Article contains additional citation context not shown here]

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a userlevel, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

....by the C interface is minimal because there is no extra data copying, nor is the function call overhead significant, particularly when compared to the granularity of communication routines in message passing architectures. For single node performance, we have tested various prototype LAPACK (Dongarra et al. 1993) modules on several architectures and found that they achieve competitive performance with similar optimized Fortran LAPACK routines. Figure 4, for example, illustrates the performance of the LU factorization routine on an IBM RS 6000 Model 550 workstation. Three versions are compared: the native ....

....researchers from universities, government laboratories, and industry. The CRPC sponsored the first workshop leading to the development of the MPI draft standard in April 1992 (Walker 1992) and a preliminary draft proposal was put forward by Dongarra, Hempel, Hey, and Walker to foster discussion (Dongarra et al. 1993). A standard message passing interface is a key component in building a concurrent computing environment in which applications, software libraries, and tools can be transparently ported between different machines. MPI provides a number of features that are useful in the design of parallel software ....

Dongarra, J. J., R. Hempel, A. J. G. Hey, and D. W. Walker (1993, February). A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory.

....any contention as shown in Figure 23 (b) In general, each column of the template has P=GCD root nodes in a stage, which broadcast their blocks of B over GCD processors of the column, where GCD denotes the greatest common divisor of P and Q. These operations are a form of group communication [15]. For A T Delta B in Section 3.3, the partial products in the same column of the processors are combined and the sum is stored in the root (destination) node. A special collecting scheme has been developed for the Delta to avoid network contention. The new collecting scheme on a linear array ....

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM12231, Oak Ridge National Laboratory, March 1993.

....its implementation on parallel architectures, particularly for data parallel computation. It includes directives for expressing data distribution across multiple memories, and concurrent execution features. The Message Passing Interface (MPI) defines a set of library routines for message passing [11], and is intended to become a widely used standard for writing parallel programs with explicit message passing, gradually replacing the vendor specific and other interfaces, which are in use today. 5 Conclusions We have argued in this paper that meaningful comparisons across DM systems require ....

J. J. Dongarra, R. Hempel, A. J. G. Hey and D. W. Walker. A proposal for a userlevel, message-passing interface in a distributed memory environment, Technical Report TM-12231, Oak Ridge National Laboratories, October, 1992.

....message passing interface were discussed, and a working group established to continue the standardization process. A preliminary draft proposal, now known as MPI 0, was put forward by Dongarra, Hempel, Hey, and Walker in November 1992, and a revised version was completed in February 1993 [3]. MPI 0 embodies the main features that were identified at the Williamsburg workshop as being necessary in a message passing standard. This proposal was intended to initiate discussion of standardization issues within the distributed memory concurrent computing community, and has served as a basis ....

....in this example. Finally, the migrating particles are communicated by a call to MPI Sendrecv. The offsets in the Ztype datatype are interpreted relative to the address of the start of the particle array. struct Pstruct particle[1000] recvbuf[100] MPI Datatype Ptype, Ztype; MPI Datatype Stype[3]=fMPI Double, MPI Double, MPI Intg; MPI Aint Sindex[3] int Sblock[3] f1, 1, 1g; int Pindex[100] int Pblock[100] Sindex[0] 0; Sindex[1] sizeof(double) Sindex[2] 2 sizeof(double) MPI Type struct (3, Sblock, Sindex, Stype, Ptype) j=0; for (i=0;i 1000;i ) if (particle[i] x right edge) ....

[Article contains additional citation context not shown here]

J. J. Dongarra, R. Hempel, A. J. G. Hey, and D. W. Walker. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report TM-12231, Oak Ridge National Laboratory, February 1993.

No context found.

Dongarra, J. J., Hempel, R., Hey, A. J. G., and Walker, D. W. A proposal for a user-level, message passing interface in a distributed memory environment. Technical Report, TM-12231, Oak Ridge National Laboratory, Available at http://www.epm.ornl.gov/~walker/mpi1.ps, February 1993.

No context found.

J. Dongarra, R. Hempel, A. Hey, and D. Walker. A proposal for a user-level, messagepassing interface in a distributed memory environment. Technical Report TM-12231, ORNL, June 1993.

No context found.

J. Dongarra, R. Hempel, A. Hey, and D. Walker. A proposal for a user-level, message-passing interface in a distributed memory environment. Technical Report TM-12231, ORNL, June 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