Vic, Sergi, from my point of view for real High-End workloads the complete I/O stack needs to be fine tuned and well understood in order to provide a good system to the users.
- Application(s) + I/O Lib(s) + MPI + Parallel Filesystem (e.g. GPFS) + Hardware (Networks, Servers, Disks, etc.) One of the best solutions to bring your application very efficently to work with a Parallel FS is Sionlib from FZ Juelich: Sionlib is a scalable I/O library for the parallel access to task-local files. The library not only supports writing and reading binary data to or from from several thousands of processors into a single or a small number of physical files but also provides for global open and close functions to access SIONlib file in parallel. SIONlib provides different interfaces: parallel access using MPI, OpenMp, or their combination and sequential access for post-processing utilities. http://www.fz-juelich.de/ias/jsc/EN/Expertise/Support/Software/SIONlib/_node.html http://apps.fz-juelich.de/jsc/sionlib/html/sionlib_tutorial_2013.pdf -frank- P.S. Nice blog from Nils https://www.ibm.com/developerworks/community/blogs/storageneers/entry/scale_out_backup_with_tsm_and_gss_performance_test_results?lang=en Frank Kraemer IBM Consulting IT Specialist / Client Technical Architect Hechtsheimer Str. 2, 55131 Mainz mailto:[email protected] voice: +49171-3043699 IBM Germany _______________________________________________ gpfsug-discuss mailing list gpfsug-discuss at gpfsug.org http://gpfsug.org/mailman/listinfo/gpfsug-discuss
