Best practices for benchmarking I/O intensive process?

[Behrooz Nobakht]

Hi,

I am interested to be able to perform benchmarks on an application
that has intensive I/O specifically using Java File API (legacy and NIO).
Let's say that a layer can be plugged in the application that can
use a Java standard API for I/O (for example read/writing files either
directly using FileIS/FileOS or using java.util.Files API on top of Path).
So, benchmarking the application would help better understand which
layer or plugin of I/O could be useful according the underlying setup
(that can involve different file systems such as NFS/XFS/etc).

I am looking for possibly best practices on how to use JMH for this?

Thanks,
Behrooz