Hi, I'm trying to test random read performance for a specific workload with a set of precreated files of various sizes using the opendir option. When increasing the number of jobs, I get performance results that are way too high for the device I'm testing. I assume this is due to caching either at the host level (although my test set is almost 3 times as large as the available ram) or in the device itself. In this case, I believe the results would be more reliable if fio would use a different (precreated) input directory for every job clone. I tried using: opendir=/mountpoint/of/device/$jobnum But that syntax is currently not supported (tested using fio version fio-3.12-3-g46ff7). Is there another way I can test random read performance with e.g. numjobs=16 and be sure that no single file is read by more than one job clone? Best regards, Filip Vervloesem
