The Experts below are selected from a list of 15 Experts worldwide ranked by ideXlab platform
Edgar Gabriel - One of the best experts on this subject based on the ideXlab platform.
-
evaluating algorithms for shared File Pointer operations in mpi i o
International Conference on Computational Science, 2009Co-Authors: Ketan Kulkarni, Edgar GabrielAbstract:MPI-I/O is a part of the MPI-2 specification defining File I/O operations for parallel MPI applications. Compared to regular POSIX style I/O functions, MPI I/O offers features like the distinction between individual File Pointers on a per-process basis and a shared File Pointer across a group of processes. The objective of this study is the evaluation of various algorithms of shared File Pointer operations for MPI-I/O. We present three algorithms to provide shared File Pointer operations on File systems that do not support File locking. The evaluation of the algorithms is carried out utilizing a parallel PVFS2 File system on an InfiniBand cluster and a local ext3 File system using a 8-core SMP.
David J Dewitt - One of the best experts on this subject based on the ideXlab platform.
-
spiffi a scalable parallel File system for the intel paragon
IEEE Transactions on Parallel and Distributed Systems, 1996Co-Authors: Craig S Freedman, Josef Burger, David J DewittAbstract:This paper presents the design and performance of SPIFFI, a scalable high-performance parallel File system intended for use by extremely I/O intensive applications including "Grand Challenge" scientific applications and multimedia systems. This paper contains experimental results from a SPIFFI prototype on a 64 node/64 disk Intel Paragon. The results show that SPIFFI provides high performance and linear scaleup on real hardware. The paper also explains how shared File Pointers (i.e., File Pointers that are shared by multiple processes) can simplify the design of a parallel application. By sequentializing I/O accesses and by providing dynamic I/O load balancing, a shared File Pointer may even improve an application's performance. This paper also presents the predictions of a SPIFFI simulator that we validated using the prototype. The simulator results show that SPIFFI continues to provide high performance even when it is scaled to configurations with as many as 128 disks or 256 compute nodes.
Ketan Kulkarni - One of the best experts on this subject based on the ideXlab platform.
-
evaluating algorithms for shared File Pointer operations in mpi i o
International Conference on Computational Science, 2009Co-Authors: Ketan Kulkarni, Edgar GabrielAbstract:MPI-I/O is a part of the MPI-2 specification defining File I/O operations for parallel MPI applications. Compared to regular POSIX style I/O functions, MPI I/O offers features like the distinction between individual File Pointers on a per-process basis and a shared File Pointer across a group of processes. The objective of this study is the evaluation of various algorithms of shared File Pointer operations for MPI-I/O. We present three algorithms to provide shared File Pointer operations on File systems that do not support File locking. The evaluation of the algorithms is carried out utilizing a parallel PVFS2 File system on an InfiniBand cluster and a local ext3 File system using a 8-core SMP.
Carlos R Vanegas - One of the best experts on this subject based on the ideXlab platform.
-
a framework architecture for shared File Pointer operations in open mpi
2013Co-Authors: Carlos R VanegasAbstract:MPI is a message passing interface that provides a distributed memory programming model for parallel computation. MPI-I/O is a parallel I/O library that is part of the MPI-2 specification. As an intermediate layer between the application and the File system, MPI-I/O is able to support features not directly enabled in the underlying File system. One such feature is a shared File Pointer. A shared File Pointer is a File Pointer to an open File that is shared between the processes that opened the File. The objective of this thesis is to develop a framework for shared File Pointer operations in OMPIO for MPI-I/O, implement and evaluate various existing and new algorithms for shared File Pointers, and to develop and evaluate a selection logic to decide which algorithm to use. Four algorithms are implemented in this thesis: the locked File, the shared memory, the individual File, and the additional process algorithms. The results show that the shared memory algorithm is the fastest. Unfortunately, it can only be used if all of the processes are executing on a single node. The individual algorithm provides a good option when running on multiple nodes, but only supports write operations. The additional process algorithm can be run from multiple nodes and can be used on all File systems, but may not be supported on all environments due to the requirement of spawning an additional process.
Craig S Freedman - One of the best experts on this subject based on the ideXlab platform.
-
spiffi a scalable parallel File system for the intel paragon
IEEE Transactions on Parallel and Distributed Systems, 1996Co-Authors: Craig S Freedman, Josef Burger, David J DewittAbstract:This paper presents the design and performance of SPIFFI, a scalable high-performance parallel File system intended for use by extremely I/O intensive applications including "Grand Challenge" scientific applications and multimedia systems. This paper contains experimental results from a SPIFFI prototype on a 64 node/64 disk Intel Paragon. The results show that SPIFFI provides high performance and linear scaleup on real hardware. The paper also explains how shared File Pointers (i.e., File Pointers that are shared by multiple processes) can simplify the design of a parallel application. By sequentializing I/O accesses and by providing dynamic I/O load balancing, a shared File Pointer may even improve an application's performance. This paper also presents the predictions of a SPIFFI simulator that we validated using the prototype. The simulator results show that SPIFFI continues to provide high performance even when it is scaled to configurations with as many as 128 disks or 256 compute nodes.