I am running iozone tests to verify interfaces on my board...

From stats I have gathered I have found that when I specify '-I' option... (meaning to perform test with direct I/O)

I get lesser read and write speed compared to simpler (without '-I' option = file system caches enabled) tests...

The problem I am facing is the scenario of having lesser read speed for direct I/O and more for cache enabled I/O is true for all the interfaces

BUT

when I look at write speed stats I found that SATA interface (harddisk) stats show better write speed for direct I/O compared to simpler (file system cache enabled) one. wherreas all other interfaces USB, MMC have beter write speed for cache enabled I/O

Can anyone please figure out, how the heck this happens???