H. Shan, J. P. Singh, R. Biswas, and L. Oliker. A Comparison of Three Programming Models for Adaptive Applications on the Origin2000. Proc. SC'2000.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in the modern computers this characteristic might have a negative impact on performance and scalability. Various techniques, such as code restructuring to increase data reuse and introducing blocking in data accesses, can address the problem and yield performance competitive with message passing [Singh], however at the cost of compromising the ease of use feature. Distributed memory models such as message passing or onesided communication offer performance and scalability but they compromise the ease of use. In this context, the message passing model is sometimes referred to as assembly ....

J.P. Singh, A Comparison of Three Programming Models for Adaptive Applications on the Origin2000, Proc. SC2000.

....memory programming paradigm on distributed memory systems is TreadMarks [2] Comparisons of the TreadMarks systems with message passing programming are given in [7] and [11] There area number of papers reporting on comparisons of different programming paradigms. As an example we name [15] and [16] where message passing and shared memory programming are compared on shared memory architectures. 7 Conclusions and Future Work We have measured the performance of OpenMP DSM implementations of three of the NAS Parallel Benchmarks on a commodity cluster of PCs and we compared the speed up to ....

H. Shan and J. Pal Singh. Comparison of Three Programming Models for Adaptive Applications on the Origin 2000.

....maintain a notion of simulation time and for many applications, the communication and computation activity for each timestep is static. For this reason, we focus our measurements on the activity for one timestep of each application. For those applications that have changing communication patterns [20], such as adaptive mesh refinement, we pay special attention, and report the communication characteristics for several different timesteps of the application. 3 Applications For our investigation, we targeted a substantial number of very sophisticated scientific applications. Table 1 provides ....

Shan, H., J.P. Singh et al., "A Comparison of Three Programming Models for Adaptive Applications on the Origin2000.

....citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and or a fee. SC2001 November 2001, Denver c 2001 ACM 1 58113 293 X 01 0011 5. 00 eral application domains indicate interesting trends and trade off s [4, 5, 16, 19]. Message passing tends to outperform shared memory, by giving the programmer opportunities to algorithmically minimize the communication overhead. The performance margin between the two programming models is not prohibitive for using shared memory though. On the contrary, it has been shown that ....

....is not prohibitive for using shared memory though. On the contrary, it has been shown that with a handful of manual optimizations for improving memory access locality and load balancing, shared memory programming models can approximate the performance of message passing with less coding effort [19]. Moving one step further, work from the authors has shown that it is possible to use flat directive based shared memory parallelism without explicit interfaces for thread or data placement, and yet be able to sustain performance as good as that of message passing or data parallel models, using ....

[Article contains additional citation context not shown here]

H. Shan, J. P. Singh, R. Biswas, and L. Oliker. A Comparison of Three Programming Models for Adaptive Applications on the Origin

....the parallel computing community usually considers two models: message passing and shared memory. Both models usually require domain partitioning and load balancing. For dynamic, adaptive applications this effort can amount to 25 of the entire code and become a significant source of overhead [9] [12]. Message passing in addition requires distributing data structures across processors and explicitly handling inter processor communication. Performance also decreases for fine grained parallelism under both models, as the effects of synchronization and communication overhead become a bigger ....

....approach is more than twice as fast as the traditional approach. Fig. 4. XMT vs. SPLASH implementation for radix sort. Speedups of parallel versions over the serial version on increasing number of TCUs. 6 Related Work Recent work on comparing different parallel programming models [9] 5] [12], 4] typically focuses on the shared memory and message passing programming models on multiprocessor systems. Our work attempts to examine parallel programming with respect to the different assumptions implied by an on chip environment. Various other projects explore on chip parallel ....

H. Shan, J.P. Singh, L. Oliker, R. Biswas, "A Comparison of Three Programming Models for Adaptive Aplications on the Origin2000," Proc. of Supercomputing (SC), 2000.

....supercomputers. The unstructured, dynamic nature of many systems worth simulating, however, makes their efficient parallel implementation a daunting task. This is primarily due to the load imbalance created by the dynamically changing nonuniform grids and the irregular data access patterns [15, 16, 22]. These cause significant communication at runtime, leaving many processors idle and adversely affecting the total execution time. Furthermore, modern computer architectures, based on deep memory hierarchies, show acceptable performance only if users care about the proper distribution and ....

....load balance and maintain good cache locality for adaptive applications. Unfortunately, a significant overhead is generally associated 1 The tuple fx; y; zg denotes fSMP nodes, MPI tasks, OpenMP threadsg. 16 L. OLIKER, X. LI, P. HUSBANDS, AND R. BISWAS with these rebalancing phases [15, 16, 22]. The CC NUMA and MPI OpenMP strategies would thus be comparable to an MPI implementation, requiring similar amounts of programming effort and rebalancing overheads. The major difference would be the use of a shared address space (global on an Origin2000, local within a node on a SP3) instead of ....

H. Shan, J. P. Singh, L. Oliker, and R. Biswas, A comparison of three programming models for adaptive applications on the Origin2000, in Proc. Supercomputing'00, Dallas, TX, 2000.

No context found.

H. Shan, J. P. Singh, R. Biswas, and L. Oliker. A Comparison of Three Programming Models for Adaptive Applications on the Origin2000. Proc. SC'2000.

No context found.

Shan H., Singh J.P., Oliker L. and Biswas R., "A Comparison of Three Programming Models for Adaptive Applications on the Origin2000", in Supercomputing 2000.

No context found.

H. Shan, J.P. Singh, L. Oliker, R. Biswas, "A Comparison of Three Programming Models for Adaptive Aplications on the Origin2000," Proc. of Supercomputing (SC), 2000.

No context found.

H. Shan, J.P. Singh, L. Oliker, R. Biswas, "A Comparison of Three Programming Models for Adaptive Aplications on the Origin2000," Proc. of Supercomputing (SC), 2000.

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