memory-eater doesn't seem to be properly measuring time when run on a device using a swap file. When using a swap file, the emmc is fully utilized and it is spending a lot of time waiting for io operations, but this doesn't seem to be factored into the time.

What steps will reproduce the problem?
(1) run the hardware_MemoryZRAMThroughput autotest or run memory-eater on a device using a swap file

What is the expected result?
The page access throughput (using 30% swap if running the autotest)

What happens instead?
The output numbers don't make sense given how long the test is taking to run and the expected performance of the device.