Steven A. Moyer and V. S. Sunderam. Scalable concurrency control for parallel file systems. In Ravi Jain, John Werth, and James C. Browne, editors, Input/Output in Parallel and Distributed Computer Systems, volume 362 of The Kluwer International Series in Engineering and Computer Science, chapter 10, pages 225--244. Kluwer Academic Publishers, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....throughput I O, but these systems are almost exclusively tailored for the sort of structured I O patterns associated with regular scientific applications. Nieuwejaar and Kotz s extensions to I O paradigms to allow e#cient strided access [NK96] are a good example of this line of work, as are [MS96] and [DJT96] The recent availability of cheap symmetric multiprocessor (SMP) systems has helped to shift the focus of parallel computing. Many desktop workstations now have two or four processors and run traditional POSIX compliant operating systems instead of the highly specialized ....

....programs. MPI IO is an extension to MPI that defines I O primitives appropriate for these sorts of parallel applications. It is highly geared for structured scientific applications, and it includes many pieces of other parallel I O frameworks, including PPFS [HER 95] PIOUS [MS94, MS96] and Vesta [CF94] MPI IO is targeted at scientific applications, and it specifically favors performance over functionality [CFF 96] It does not attempt to be compatible with the UNIX (POSIX) file interface, which is acceptable because programs must generally 31 be rewritten from scratch ....

Steven A. Moyer and V. S. Sunderam. Scalable concurrency control for parallel file systems. In Ravi Jain, John Werth, and James C. Browne, editors, Input/Output in Parallel and Distributed Computer Systems, volume 362 of The Kluwer International Series in Engineering and Computer Science, chapter 10, pages 225--244. Kluwer Academic Publishers, 1996.

....example showing the difficulties of moving from the sequential to the parallel I O domain is discussed. Section 4 presents the parallel I O templates proposed by this paper. Section 5 compares the conversion steps in developing a parallel application against an existing parallel I O system (PIOUS [27]) Section 6 presents the actual performance of two applications. The first application is the one presented in Section 5 which has fine grained I O spread throughout the application. The second application uses large grained I O operations. Finally, the conclusions are presented. 2 Related Work ....

....35] These studies look at traces of actual parallel applications doing I O using specific parallel I O libraries and architectures. Characterizing well understood parallel programs under controlled conditions facilitates development of optimization techniques. Parallel I O systems such as PIOUS[27], MPI IO [8] and ELFS [19] have abstracted parallel I O semantics so that there are two types of I O streams sequential and parallel. Additionally, these I O systems are designed to use the parent communication system (PVM [16] MPI [37] and Mentat [18] respectively) This approach of ....

[Article contains additional citation context not shown here]

S. A. Moyer and V. S. Sunderam, "Scalable Concurrency Control for Parallel File Systems," In proceedings of Third Annual Workshop on Input/Output in Parallel and Distributed Systems, pp. 90-106, Santa Barbara, CA, 1995.

....discussing the performance results. 1. Introduction The development of parallel applications has focused on computational parallelism, with the result that parallel I O implementation techniques have lagged behind. Support for parallel I O is typically obtained from library packages such as PIOUS [8] or MPI IO [2] While these libraries support a common and standardized interface respectively, they still require that the developer explicitly differentiate between parallel and sequential I O streams at the source code level, and often require that application data be imported or exported to ....

S. Moyer and V. Sunderam, "Scalable Concurrency Control for Parallel File Systems," Third Workshop on I/O in Parallel and Distributed Systems, pp. 90-106, 1995.

....applicability to the execution of parallel applications: a segmented concurrent write benchmark (SCWB) and a interleaved concurrent write benchmark (ICWB) SCWB uses a parallel program with partitioned access similar to the BLOCK (2,0,0,0) of figure 5. This benchmark, similar to the described in [Moyer95], divides a file into contiguous segments, one per process, with each segment accessed sequentially by a different process. ICWB uses a parallel program, where each process writes the file in a interleaved fashion, similar to the CYCLIC (1,4,0,0) of the figure 5. The parallel program for each ....

S. A. Moyer and V. S. Sunderam. Scalable concurrency control for parallel file systems. Technical Report CSTR-950202, Department of Math and Computer Science, Emory University, Atlanta, GA 30322, USA, 1995.

....Integration with the computational parallelism and the necessary information needed by the PPS to execute the parallel I O behaviour is examined. Section 6 compares the conversion steps in developing a parallel application using an implementation of PI OT and an existing parallel I O system (PIOUS [28]) Section 7 presents the actual performance of two applications: one with fine grained I O and one large grained I O access patterns. Finally, the conclusions are presented. 4 2. Related Work Templates, or predefined behaviours, have been applied to the expression of computational ....

....using specific parallel I O libraries and architectures. Characterizing well understood parallel programs under controlled conditions facilitates the development of optimization techniques. We have used the results of these studies to drive our template design. Parallel I O systems such as PIOUS[28], MPI IO [8] and ELFS [19] have abstracted parallel I O semantics so that there are two types of I O streams sequential and parallel. Additionally, these I O systems are designed to work with the parent communication or PPS system (PVM [16] MPI [38] and Mentat [18] respectively) The ....

[Article contains additional citation context not shown here]

S. A. Moyer and V. S. Sunderam, "Scalable Concurrency Control for Parallel File Systems," In proceedings of Third Annual Workshop on Input/Output in Parallel and Distributed Systems, pp. 90-106, Santa Barbara, CA, 1995.

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