O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping, The Paderborn University BSP(PUB) Library-Design, Implementation and Performance, IPPS/SPDP'99, April 1999, 99--104.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....and available for wide range of hardware including our available hardware, a Sun multiprocessor, we use the BSPlib to express BSP implementation template and to write real BSP programs to which example problems of VEC BSP are hand compiled. Recently, the Paderborn University BSP (PUB) Library [15] has been developed. The PUB library offers the same functionality as BSPlib, but in addition provides several other features. In particular, it has a mechanism to partition the machine into subsets which synchronise independently. In this way more complex programs made of patterns which ....

O. Bonorden, B. Juulink, I. von Otto, and I. Rieping. The Paderborn University BSP (PUB) Library-design, Implementation and Performance. In 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing, April 1999.

....several library packages with a C interface for the BSP model. Processors interact by two sided message passing or by one sided communication [direct remote memory access (DRMA) collective communication is supported. The two most widely known BSP libraries are Oxford BSPlib [6] and Paderborn PUB [2], where the latter also supports dynamic splitting of processor groups. According to our knowledge, NestStep is the first proper programming language for the BSP model. HPF [5] is a SIMD language and provides, at least from the programmer s point of view, a sequentially consistent shared memory. ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library: Design, Implementation and Performance. In Proc. IPPSSPDP'99 IEEE Int. Parallel Processing Symp. and Symp. Parallel and Distributed Processing, 1999.

....architectures; once again, implementations of OpenMP are available both in open source and from commercial vendors. There are also several higher level parallel programming abstractions that use MPI, OpenMP, or POSIX threads, such as implementations of the Bulk Synchronous Parallel (BSP) model [75, 42, 22] and data parallel languages like High Performance Fortran [41] Higher level application framework such as KeLP [29] and POOMA [27] also abstract away the details of the parallel communication layers. These frameworks enhance the expressiveness of data parallel languages by providing the user ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) library - design, implementation and performance. In Proc. 13th Int'l Parallel Processing Symp. and the 10th Symp. Parallel and Distributed Processing (IPPS/SPDP), San Juan, Puerto Rico, 1999. www.uni-paderborn.de/~pub/.

....to separate the communication diculties from the scheduling diculties. It is becoming more and more popular for theoretical studies, providing a solid foundation for designing parallel algorithms. Moreover, some implementations of BSP library components have been developed on several platforms [16, 3]. 2.2.1 De nition Programming in BSP consists of a succession of supersteps. Each superstep may be divided into three activities: a computation phase where independent local computations run in parallel; a communication phase where the communications of data involve all the processors in ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The paderborn university bsp (pub) library - design, implementation and performance. In Proc. of 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San Juan, Puerto Rico, April 1999. url:

....architectures; once again, implementations of OpenMP are available both in open source and from commercial vendors. There are also several higher level parallel programming abstractions that use MPI, OpenMP, or POSIX threads, such as implementations of the Bulk Synchronous Parallel (BSP) model [77, 43, 22] and data parallel languages like High Performance Fortran [42] Higher level application framework such as KeLP [29] and POOMA [27] also abstract away the details of the parallel communication layers. These frameworks enhance the expressiveness of data parallel languages by providing the user ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) library - design, implementation and performance. In Proc. 13th Int'l Parallel Processing Symp. and the 10th Symp. Parallel and Distributed Processing (IPPS/SPDP), San Juan, Puerto Rico, 1999. www. uni-paderborn.de/pub/.

....allowing a process to manipulate certain registered areas of a remote memory which have been previously made available by the corresponding processes. Second, we have also developped codes for the send receive paradigm that is also avalaible in BSP (it is more developped with the PUB distribution [9]) The main 4 disadvantage of using DRMA primitives is that the register step can be very time consuming (see addendum) in particular with BSPLib (see [1] for an example) These two programming style allow us to compare the eciency of communications. DRMA has the advantage that programmer deal ....

I. von Otte O. Bonorden, B. Juurlink and I. Rieping, \The paderborn university bsp (pub) library - design, implementation and performance", in 13th International Parallel Processing Symposium and 10th Symposium on Parallel and Distributed Processing, 12 - 16 April,

....should be an environment on parallel machines that supports the coding of BSP algorithms independent from the underlying machine and that executes the code efficiently. Such BSP environments are the Oxford BSPlib [10] the Green library [9] and the PUB Lib (Paderborn University BSP Library) [5]. The PUB Lib is a C Library to support the development and implementation of parallel algorithm designed for the BSP model. It provides the use of block wise communication as suggested by the BSP model, and it provides the use of locality as suggested in the D BSP model by allowing to ....

O. Bonorden, B. Juurlink, I. von Otte, I. Rieping. The Paderborn University BSP (PUB) Library --- Design, Implementation and Performance. In: Proc. 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), 1999, pp. 99--104.

....This yields an expression of the form a bg cl. For further details and some basic techniques, see [6] BSPlib [21] is a standard library which enables parallel programming in BSP style. Available implementations are the Oxford BSP toolset [22] and the Paderborn University BSP library [8]. 3.2. Data structures and data distributions. At each stage k, 1 # k # log 2 N M , the number of intermediate polynomial pairs doubles as the number of expansion coe#cients halves. Thus, at every stage of the computation, all the intermediate polynomials can be stored in two arrays of size ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping, The Paderborn University BSP (PUB) library---design, implementation and performance, in Proceedings of the 13th International Parallel Processing Symposium and the 10th Symposium on Parallel and Distributed Processing, CD-ROM, IEEE Computer Society, Los Alamitos, CA, 1999.

.... devising cost functions for other network architectures, such as 100Mbps or Gigabit Ethernet, Myrinet or ATM, or comparing the performance predictability levels achieved by BSPlib against those attained by other communication libraries, such as the BSP PUB library developed at Paderborn University [3]. Acknowledgments We are grateful to Nancy Amato and Andrea Pietracaprina for helping set the ground for this research. We also wish to thank the EUROPAR referees for their valuable comments and suggestions. ....

O. Bonorden, B. Juurlink, I. von Otte, and I Rieping. The Paderborn university BSP (PUB) Library -- Design, Implementation and Performance In Proc. of the 2nd merged IPPS/SPDP Symp., pages 99-104, San Juan, Puerto Rico, April 1999.

....supersteps. This yields an expression of the form a bg cl. For further details and some basic techniques, see [4, 16] The second reference describes BSPlib, a standard library defined in May 1997 which enables parallel programming in BSP style. The Paderborn University BSP (PUB) library [6] is another library that permits programming in BSP style; it provides the extra feature of subset synchronization. 1.3. Parallel radix 2 FFTs. Since the introduction of parallel computers, and even before that, methods for parallelizing FFT algorithms have been proposed [23] The earliest methods ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) library -- design, implementation and performance. In 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San Juan, Puerto Rico, April 1999.

....at most h packets, and each destination receives at most h packets. One motivation for designing algorithms that route h relations is that routing an h relation is the primitive communication step in the BSP model of computation [52] for which there are growing libraries of parallel programs [6, 21, 36, 41]. Finally, we show that a twinbutterfly whose individual splitters do not exhibit expansion can emulate a bounded degree multibutterfly with an (ff; fi) expansion property, for any ff Delta fi 1=4. The fact that an N node multibutterfly network contains an N node AKS network does not imply that ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) library -- design, implementation and performance. In Proceedings of 13th Inter37 national Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San Juan, Puerto Rico, April, 1999.

.... g(k) l(k)g where k is the size of the current submachine. The cost of the partition is the maximal cost of the created submachines. The total cost is the sum of the costs of the individual supersteps. A realization of this BSP model can be found in the Paderborn University BSP (PUB) library [3], which allows partitioning as an additional feature. The Y PRAM [16] is, like the H PRAM, a shared memory model. The p processor machine is recursively partitionable into two sub PRAMs of size p=2. There are two parameters of the machine: a latency l(k) and a bandwidth inefficiency (k) not ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library - Design, Implementation and Performance. In International Parallel Processing Symposium, IEEE Computer Society Press,

.... g(k) l(k)g where k is the size of the current submachine. The cost of the partition is the maximal cost of the created submachines. The total cost is the sum of the costs of the individual supersteps. A realization of this BSP model can be found in the Paderborn University BSP (PUB) library [3], which allows partitioning as an additional feature. The Y PRAM [16] is, like the H PRAM, a shared memory model. The p processor machine is recursively partitionable into two sub PRAMs of size p=2. There are two parameters of the machine: a latency l(k) and a bandwidth ineciency (k) not ....

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library - Design, Implementation and Performance. In International Parallel Processing Symposium, IEEE Computer Society Press,

....i.e, the phenomenon that when several processors send to the same processor at the same time, stalls occur. We also discuss how these issues are dealt with differently in two communication libraries based on the BSP model [20, 16] BSPlib [10] and the Paderborn University BSP (PUB) library [6], and shortly discuss how different models lead to different algorithmic techniques. 2 Block Transfers It is well known that on many parallel computers, there is a high startup cost associated with each message transmission. For example, in [9] it is shown that, typically, the messages must be ....

....this architecture and also of others that have a significant startup cost. The main argument is that sometimes higher performance can be achieved by increasing the communication volume, which clearly contravenes the incentives provided by the BSP model. The Paderborn University BSP (PUB) library [6] provides primitives for collective communication operations like broadcast, parallel prefix, etc. for the following reasons. First, they are used frequently in parallel applications and often determine the overall efficiency. For example, the sole communication routine in the Linpack benchmark ....

[Article contains additional citation context not shown here]

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library -- Design, Implementation and Performance. In Proc. IPPS/SPDP, 1999.

No context found.

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping, The Paderborn University BSP(PUB) Library-Design, Implementation and Performance, IPPS/SPDP'99, April 1999, 99--104.

No context found.

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library - Design, Implementation and Performance. In Proc. of 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San-Juan, Puerto-Rico, April 1999.

No context found.

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library - Design, Implementation and Performance. In Proc. of 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San-Juan, Puerto-Rico, April 1999.

No context found.

O. Bonorden, B. Juurlink, I. von Otte, and I. Rieping. The Paderborn University BSP (PUB) Library - Design, Implementation and Performance. In Proc. of 13th International Parallel Processing Symposium & 10th Symposium on Parallel and Distributed Processing (IPPS/SPDP), San-Juan, Puerto-Rico, April 1999.

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