Hi, I was about to check IOperformance with memory mapped files dependent on how many threads and how many files give best results. (For this test I used read only) The sum of all files used for this test were always 1GB. The usesd files were created whith [dd if=/dev/zero of=file bs=128k count = ... ] The block size for each IO was 16kB. The result is given in the picture I attached to this posting. As you can see there is a gain in IO/s when increasing the #threads until 8 (which is explainable as i work with an 8 CPU system). But why do i get more performance when the number of files grows bigger than 32? Can anyone explain? What influence has the number of files when using mmap?Thanks for help
My system: solaris 9 platform with an 8cpu USPARK system. If you are interested, i attached the source code for my testtool This message posted from opensolaris.org
ro_mm1.png
Description: PNG image
rand.c
Description: Binary data
_______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
