Hi, on http://www.rzuser.uni-heidelberg.de/~jkunkel2/150GB.tgz you will find a tar archive comparing ODirect disk access on a cool raid machine accessing a total precreated file with a size of 150GByte. Up to the size of 128KByte all threads iterate 10000 times and access the given blocksize randomly. Starting with 256KByte they iterate only 1000 Times. You will find average throughput and also average wallclock times per iteration. I don't want to say much about the results. O_Direct wins in all cases, even if you do a read modify write. The actual performance we should get is between the raw Read and ReadWrite performance due to the 512 alignment necessary to O_Direct I/O. Interesting for example doing small I/O O_Direct readwrite access is twice as fast as system I/O. Determing the actual strategy: caching vs no-caching and thread number will be quite complex. The patch which allows the user to increase the buffer size to dbpf / flow will be useful, too.
@Pete Are the read/write throughput of 100MByte/75MByte for read and write below the raid systems capabilities ? Should be more like 150 MByte ? There are a lot of thinks which might help... Thanks, Julian _______________________________________________ Pvfs2-developers mailing list [email protected] http://www.beowulf-underground.org/mailman/listinfo/pvfs2-developers
