R.W. Hockney and E.A. Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing 18 (1992) 1067-72.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... 3.1.11 PowerPC model 250T : Tci17 is a 66MHz PowerPC RISC 6000 model 250T (approx. 50 MFlops) Compilation options xlf O3 qhot qnosave qarch=ppc 3. 2 Message passing performance The performance of the message passing system can be characterised using the parameters of Hockney and Carmona [21], namely ffl r 1 , the asymptotic transfer rate in MB s and ffl n 1 2 , the message length at which half this rate is achieved. A variety of message passing harnesses have been investigated, including MPL, MPI, PVM3, PVMe, TCGMSG, GA Tools, HPF and BSP. MPICH (Argonne National Laboratory, ....

R.W. Hockney and E.A. Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing 18 (1992) 1067-72.

....obtained a mean performance of approximately 20.62 Mflops when using BLAS 3, and 7.29 Mflops when using BLAS 1. In this way, the values of k 3 and k 1 are 0.048 and 0.137 s, respectively. The values of the start up time and the word sending time can be approximated by 61 and 0. 149 s, respectively [15]. With these data the expected and real times obtained with different number of processors and the matrix sizes of table 9, are shown in table 10. The expected times have been obtained using formulas 7 and 8 for the sequential algorithm, and formulas 20 and 21 for the parallel algorithm. Also the ....

R. W. Hockney and E. A. Carmona. Comparison of communications on the Intel iPSC/860 and Touchstone Delta. Parallel Computing, 18:1067-- 1072, 1992.

....are possible with a lower annex setup cost. This study is a beginning, more work needs to be done to determine the right set of network interfaces which can provide robust messaging performance. 7 Related work While several studies have measured communication performance on parallel machines [7, 21, 15], they have not explored the relationship between network interface architecture and messaging performance. A great deal of attention has also been focused separately on specialized hardware support for messaging, the design of efficient messaging layers, and network management for improving ....

R. W. Hockney and E. A. Carmona. Comparison of communication on the Intel iPSC/860 and Touchstone Delta. Parallel Computing, (18):1067--1072, 1992.

....are reported. A full description of these tests are presented in a separate report [4] A poster about the implementation of MPI on the T3D at EPCC is presented here by Gordon Smith. At this point we merely summarise these results through the tables below, which collate the Hockney and Carmona [19] parameters for the different software and networks investigated. ffl r 1 , the asymptotic transfer rate in MB s and ffl n 1 2 , the message length in Bytes at which half this rate is achieved. These parameters will be used throughout this talk in our discussion of the performance of each ....

R.W.Hockney and E.A.Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing 18 (1992) 1067-72.

....have matured, they have added multitasking operating systems and more stringent error checking which have increased the overhead associated with starting a communication. Several researchers have noted that startup cost is the predominate factor in determining the total cost of communication [10] [17] For this reason we will assume that overhead induced by limitation in actual bandwidth on the communi cation channels and link congestion are actually second order effects, and we will not consider them in our model. This makes the model much simpler, since it does not have to deduce the ....

R. W. Hockney and E. A. Carmona. Comparison of communications on the intel ipsc/860 and touchstone delta. Parallel Computing, 18(9):1067 -- 1072, 1992.

....6.9 PowerPC model 250T Tci17 is a 66 MHz PowerPC RISC 6000 model 250T (approx. 50 MFlops) Compilation options xlf O3 qhot qnosave qarch=ppc 7 Message passing performance The performance of the message passing system can be characterised using the parameters of Hockney and Carmona [36], namely ffl r 1 , the asymptotic transfer rate in MB s and ffl n 1 2 , the message length in Bytes at which half this rate is achieved. A variety of message passing harnesses have been investigated, including MPL, MPI, PVM3, PVMe, TCGMSG, GA Tools, HPF and BSP. MPICH (Argonne National ....

R.W.Hockney and E.A.Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing 18 (1992) 1067-72.

....once a week, by means of a locally developed booking system. Chapter 2 Message Passing, Languages and Mathematical Libraries 2.1 Summary of Message passing Performance R.J. Allan The performance of the message passing system can be characterised using the parameters of Hockney and Carmona [21], namely ffl r 1 , the asymptotic transfer rate in MB s and ffl n 1 2 , the message length at which half this rate is achieved. A variety of message passing harnesses have been investigated, including MPL, MPI, PVM3, PVMe, TCGMSG, GA Tools, HPF and BSP. MPICH (Argonne National Laboratory, USA) ....

R.W. Hockney and E.A. Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing 18 (1992) 1067-72.

....entry for the worst speedup prize. Something is clearly going terribly wrong when this benchmark is run on the Intel iPSC 860 and Touchstone Delta. Since these computers both use the same arithmetic chip (the original i860) but have quite different communication networks, hardware and properties [14], the almost identical bad behaviour of the two computers suggests that the problem arises in the arithmetic and not in the communications. It was therefore imperative to discover, by using performance tools, what was going wrong, and this is where Paragraph proved its usefulness. 5.1.3. ParaGraph ....

....between two neighbouring nodes. The transfer time is in fact half of the total time, measured by the benchmark for different message lengths. The measurements were taken from PARMACS, which results in about 4 10 overhead in the start up time if comparing it with the native Intel communications [14]. This is actually the price of portability on iPSC 860, which one should pay in order to have a portable program across different distributed memory computers using PARMACS macros. Two series of experiments were performed regarding the performance monitoring overhead one with and one without a ....

R.W. Hockney and E.A. Carmona, Comparison of communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing, to appear (1992).

....However, for messages of 100 byte or less Computer Benchmarks, North Holland, 1993. 11 Figure 2: The time for a single message transfer, as a function of the message length on the ipsc 860 and the Touchstone Delta the start up time was only 85 s (see Fig. 2) Using native Intel communications [11] the start up time on the ipsc 860 was measured as 205 s for long messages and 76 s for short messages. These results show a relatively small overhead of about 4 10 in the communications start up time for the parmacs macros implementation on the ipsc 860. No difference was measurable in the ....

R.W. Hockney and E.A. Carmona. Comparison of Communications on the Intel iPSC/860 and Touchstone Delta, Parallel Computing, 18, to appear (1992).

No context found.

R. W. Hockney and E. A. Carmona. Comparison of Communications on the Intel iPSC/860 and Touchstone Delta. Parallel Computing, 18:1067--1072, 1992.

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