M. Banikazemi, J. Sampathkumar, S. Prabhu, D.K. Panda, and P. Sadayappan. Communication modeling of heterogeneous networks of workstations for performance characterization of collective operations. In HCW'99, the 8th Heterogeneous Computing Workshop, pages 125--133. IEEE Computer Society Press, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....occurs in distinct phases which are serialized. The Jacobi communication phase involves a series of p broadcasts per iteration; each machine in the computation is the root of one of these broadcasts. In the MPI implementation we used in this work, the broadcast is implemented as a binomial tree [4]. As a rst approximation to modeling this communication structure, we calculate the average message time, msgTime avg , and calculate the communication time on processor P i as commTime i = p log 2 (p) msgTime avg : 5) Our Game of Life and Jacobi communication models each depend on a model ....

Banikazemi, M., Sampathkumar, J., Prabhu, S., Panda, D. K., and Sadayappan, P. Communication modeling of heterogeneous networks of workstations for performance characterization of collective operations. In Proceedings of the 8th Heterogeneous Computing Workshop (April 1999).

....) return (commTime i ) Figure IV.4: Game of Life communication cost calculation. composed of individual MPI Send and MPI Recv calls. The broadcast begins at the root node (the root is whichever processor is the initiator of the broadcast) and is sent to all other processors via a binomial tree [5]. The binomial tree broadcast structure is designed to minimize the number of serialized messages that must proceed before the broadcast is complete; theoretically at most log 2 (p) messages are serialized in each broadcast. Figure IV.5 illustrates a binomial tree broadcast structure for seven ....

....irregular data partitions, each processor s broadcast could be of a di#erent size. Additionally, since the root node will be di#erent for each broadcast in an iteration, the connections involved in the broadcast will vary from broadcast to broadcast. One modeling approach is presented in [5]; in this work the authors propose directly calculating the cost of each path in the binomial broadcast tree to determine the longest path, which is then taken to be the predicted broadcast time. This methodology has not been tested on heterogeneous, wide area resources sets and it does not ....

Mohammad Banikazemi, Jayanthi Sampathkumar, Sandeep Prabhu, Dhabaleswar K. Panda, and P. Sadayappan. Communication modeling of heterogeneous networks of workstations for performance characterization of collective operations. In Proceedings of the 8th Heterogeneous Computing Workshop, April 1999. 42, 42

....to develop optimized communication patterns. Bhat, Raghavendra, and Prasanna [BRP99] extend the FNF algorithm [BMP98] and propose several new heuristics for collective operations. Their heuristics consider the effect communication links with different latencies have on a system. Banikazemi [BSP99] present a model for point to point communications in heterogeneous networks of workstations and use it to study the effect of heterogeneity on the performance of collective operations. 75 5.1 The HBSP Programming Library The HBSP 1 collective communication algorithms are implemented using the ....

M. Banikazemi, J. Sampathkumar, S. Prabhu, D. Panda, and P. Sadayappan. "Communication Modeling of Heterogeneous Networks of Workstations for Performance Characterization of Collective Operations. " In Heterogeneous Computing Workshop (HCW '99), pp. 125-- 133, April 1999.

.... and his colleagues have surveyed collective communications on hypercubes, meshes, and tori in wormhole routed networks [90] Recently, Banikazemi and others, have proposed efficient broadcasting and multicasting algorithms using communication capabilities of heterogeneous networks of workstations [15]. In the context of optical interconnection networks, Berthome and Ferreira [20, 21] have presented broadcasting and multicasting algorithms for networks using optical passive stars (OPS) Comparative Study of one to many wavelength division multi Broadcast Scatter Gather Multinode broadcast abc ....

M. Banikazemi, J. Sampathkumar, S. Prabhu, D. K. Panda, and P. Sadayappan, "Communication Modeling of Heterogeneous Networks of Workstations for Performance Characterization of Collective Operations", Proceedings of the International Workshop on Heterogeneous Computing, in conjunction with IPPS/ SPDP'99, April 1999, pp. 125-131.

....on the FNF algorithm [3] and have proposed new heuristics for collective operations. In these heuristics, the effect of communication links with different latencies is also taken into account. A new set of performance models for collective operations in heterogeneous systems has been proposed in [4]. These models can be used to study the effect of heterogeneity on the performance of collective operations and hence design more efficient algorithms. 9 Conclusions and Future Research In this paper, we have presented three new approaches to implement fast collective communication in the ....

M. Banikazemi, J. Sampathkumar, S. Prabhu, P. Sadayappan, and D. K. Panda. Communication Modeling of Heterogeneous Networks of Workstations for Performance Characterization of Collective Operations. In Proceedings of the 1999 IPPS Heterogeneous Computing Workshop, pages 125--133, April 1999.

No context found.

M. Banikazemi, J. Sampathkumar, S. Prabhu, D.K. Panda, and P. Sadayappan. Communication modeling of heterogeneous networks of workstations for performance characterization of collective operations. In HCW'99, the 8th Heterogeneous Computing Workshop, pages 125--133. IEEE Computer Society Press, 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