IOshark is a repeatable application workload storage benchmark. You
can find more documentation on IOshark at :
https://docs.google.com/a/google.com/document/d/1Bhq7iNPVc_JzwRrkmZqcPjMvWgpHX0r3Ncq-ZsRNOBA/edit?usp=sharing

The short summary of what IOshark is : IOshark has 2 components, one
is a strace+ftrace compiler that takes straces and select ftraces fed
into it and compiles this into bytecodes (stored in *.wl files). The
compiler runs on a Linux host. The second component (which runs on the
device) is the tester that takes as input the bytecode files (*.wl
files) and executes them on the device.

How to Run :
----------
- First collect straces and compile these into bytecodes. The wrapper
script provided (collect-straces.sh) collects straces, ships them to
the host where the script runs, compiles and packages up the bytecode
files into a wl.tar file.
- Ship the wl.tar file and the iostark_bench binaries to the target
device (on /data/local/tmp say). Explode the tarfile.
- Run the tester. "ioshark_bench *.wl" runs the test with default
options. Supported ioshark_bench options :
-b : Explicitly specify a blockdev (to get IO stats from from
/proc/diskstats).
-d : Preserve the delays between successive filesystem syscalls as
seen in the original straces.
-n <N> : Run for N iterations
-t <N> : Limit to N threads. By default (without this option), IOshark
will launch as many threads as there are input files, so 1 thread/file.
-v : verbose. Chatty mode.
-s : One line summary.
-q : Don't create the files in read-only partitions like /system and
/vendor. Instead do reads on those files.