On implementing MPI-IO portably and with high performance
Proceedings of the sixth workshop on I/O in parallel and distributed systems
Automatically tuned collective communications
Proceedings of the 2000 ACM/IEEE conference on Supercomputing
MPI-IO/GPFS, an optimized implementation of MPI-IO on top of GPFS
Proceedings of the 2001 ACM/IEEE conference on Supercomputing
Data Sieving and Collective I/O in ROMIO
FRONTIERS '99 Proceedings of the The 7th Symposium on the Frontiers of Massively Parallel Computation
Collective Buffering: Improving Parallel I/O Performance
HPDC '97 Proceedings of the 6th IEEE International Symposium on High Performance Distributed Computing
STAR-MPI: self tuned adaptive routines for MPI collective operations
Proceedings of the 20th annual international conference on Supercomputing
Self-adapting numerical software and automatic tuning of heuristics
ICCS'03 Proceedings of the 2003 international conference on Computational science
Self-adapting numerical software and automatic tuning of heuristics
ICCS'03 Proceedings of the 2003 international conference on Computational science
A collective I/O implementation based on inspector---executor paradigm
The Journal of Supercomputing
Performance Evaluation of Collective Write Algorithms in MPI I/O
ICCS '09 Proceedings of the 9th International Conference on Computational Science: Part I
Extending the MPI-2 generalized request interface
PVM/MPI'07 Proceedings of the 14th European conference on Recent Advances in Parallel Virtual Machine and Message Passing Interface
Hi-index | 0.00 |
The processing of MPI-IO operations can be controlled via the MPI API using file hints, which are passed to the MPI library as MPI info objects. A file hint can affect how the MPI library accesses the file on the file system level, it can set buffer sizes, turn special optimizations on and off or whatever parameters the MPI implementation provides. However, experience shows that file hints are rarely used for reasons that will be discussed in the paper. We present a new approach which dynamically determines the optimal setting for file hints related to collective MPI-IO operations. The chosen settings adapt to the actual file access pattern, the topology of the MPI processes and the available memory resources and consider the characteristics of the underlying file system. We evaluate our approach which has been implemented in MPI/SX, NEC's MPI implementation for the SX series of vector supercomputers.