TODO

Immediate
----------

Make each SNIA test round use its own directory for its own set of files.
This will be needed to be able to later show for instance convergence
to steady state from graphs.

Medium term
-----------

CONFIG_ENERGIZED_STEADY_STATE=y is supported currently through uptime, and
assuming a file being present means the device is powered on. This is not
true. A system can be powered on for a long time, and a new driver to support
the device only added later. Addressing this seems complex and may not be
easily possible, we accept these risks today by assuming a fully deployed OS
on bootup for now, but this could use some love long term to figure out how
we actually verify this properly. If nvme drives had run time power management
enabled, they don't, we could use /sys/block/nvme0n1/power/runtime_active_time
for instance, as it reflects the amount of active seconds. If always powered
on since boot, this would match close to the value or uptime, but since this
is disabled for nvme we cannot use it. One possibility is to extract this
implementation and generalize it, so we have two values possible, one for
power run time, another for actual device power on. Some folks may end up
arguing however this may be better suited for userspace to track with udev
events.

Long term
---------

Add SNIA tests for:

  * throughput (bw)
  * latency

Consider adding the other SNIA tests.

Review if we do what we should be doing for:

  * Write cache enabled (WCE), Write cache disabled (WCD) for Enterprise and
    not Enterprise.

Graph Steady State Convergence Plot which shows general visual convergence to
Steady State by plotting the dependent variable(s) (IOPS, Throughput, etc.) for
each Round

Graph Steady State Verification Plot shows which shows via either graph or
table, that the device has reached Steady State per definition 2.1.24, by
examining dependent variable behavior within the Measurement Window.

Complete graphs for SNIA report (make graph)

Implement gathering all data as per SNIA "Common Reporting Requirements",
refer to section 5 of SSS PTS 2.0.1

Add a kconfig guard option which if enabled it will *stop* running a series
of a tests for unit tests if we've already reached a point in which the max
bandwidth / IOPS gotten worse. This should now be relatively easy since
scripts/gen-graph.sh average_column() is already implemented. It should just
be a matter of looking at the *prior* fio file ssv results, computing the
column average there and if the *current* values are worse we should indicate
this somehow on the directory with a guard file, so that on the next iteration
it skips processing the file.

Add idle measurement and visualize it on graphs somehow

Chart width to cover min and max values (min bandwidth, max) after an initial
period of ramp up time.