Disk performance `pi1` vs `pi2`/`pi3`

[carlosperate]

Mostly wondering if somebody might have looked into this, have any insights, or might know about things we could try to get better performance writing to disk with the pi2/pi3configuration?

I've noticed that most things I run inpi2 or pi3 mode are slower than in pi1. Partially this is due to the poor disk performance compared with the pi1 configuration (which is not great either; the other reason is poor network performance, but that is usually blamed on the USB ethernet device).

Running dd if=/dev/zero of=~/test.tmp bs=1M count=128 conv=fsync we find:

	dd 128 MB write
Docker host (no qemu)	0.2s
qemu versatilepb (pi1)	7 s
qemu raspi2b (pi2)	58 s
qemu raspi3b (pi3)	60 s
qemu virt *	5 s

Docker host: docker run --rm -it --entrypoint sh lukechilds/dockerpi:latest
versatilepb docker run --rm -it lukechilds/dockerpi:latest
raspi2b: docker run --rm -it lukechilds/dockerpi:latest pi2
raspi3b: docker run --rm -it lukechilds/dockerpi:latest pi3
* The virt machine type was from a different arm emulation I have in a fork of this project, which I haven't pushed to any docker registry yet, but it's here to show that other qemu machine types can also have better performance than pi2/pi3.

I haven't found much info online about the raspi2b/raspi3b machine types being this slow, as they are 8 to 12 times slower than other qemu arm emulations (all writting to an .img file in the "raw" format).
The closest I've found was this unanswered question in stackexchange, but no responses: https://raspberrypi.stackexchange.com/questions/139487/improving-file-system-performance-on-emulated-raspberry-pi