> -----Original Message----- > From: Nguyen Viet Dung [mailto:dungnguyenviet@xxxxxxxxx] > Sent: Friday, April 03, 2015 9:20 AM > To: Elliott, Robert (Server Storage) > Cc: fio@xxxxxxxxxxxxxxx > Subject: Re: Need support about value IOPS Which FIO measured > > Dear Elliot, Robert > Thanks so much yours reply. I'm not control with FIO. > So, can you suggest for my, how can i test hard disk by fio with not > buffer cache. > I sent you my command: can you show for me that fail in my command? > > sudo fio --output=/media/data/rw4k.result --ioengine=libaio -- > direct=0 --invalidate=1 --gtod_reduce=0 --name=testrw --filename=testrw > --directory=/media/data --bs=4k --iodepth=8 --size=100M --readwrite=rw - > -rwmixread=75 direct=1 will bypass the page cache, ensuring I/Os really go to the drive. size=100M is too small. If you're testing a HDD, it won't be seeking very far, and most or all the reads could easily be coming out of the drive's write cache (which could be 64 to 128 MiB). Use programs like hdparm, sdparm, and tools in the sg3_utils package to disable the drive write cache. A larger iodepth might perform better. SATA protocol supports 32, for example. SAS drives support even larger numbers. --- Robert Elliott, HP Server Storage ��.n��������+%������w��{.n�������^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�