on accuracy

[flugi]

Hello, I would like to ask how the utilized bandwidth is measured, and what are the chances that the measurements are very wrong. I am benchmarking a USB hub with 3 Kinect devices and usbtop reports ~1800KiB/s for my typical use of a single Kinect sensor. When I run three of them, the results are clearly show a bottleneck as the sum of the three bandwidth numbers usbtop show never exceeds 3000KiB/s, I see frame drops and ~15fps stream instead of 30.

However if I run a different application the same HUB and same Kinect reports 15263 KiB/s when used as webcam.

What could be the reason of these numbers? Is it possible that the actual bandwidth need by the Kinect is much more than what usbtop show?

[dfloer]

I've also been wondering something similar myself. I've got a UVC device and usbtop is only showing around 15MiB/s, which seems far too low for the amount of data it should be pushing. I'd expect more like 93MB/s for an the raw data coming from it.

[dfloer]

I did some more poking, and it seems like usbtop was accurate for the UVC device in bulk mode, but was still showing low for isochronous mode. Potentially there's a bug there.

[aguinet]

Hello everyone,

I unfortunately don't have any equipment that would allow me reproducing these experiments.

Bandwidth estimation happens in the src/usb_stats.cpp file, with this TODO: https://github.com/aguinet/usbtop/blob/master/src/usb_stats.cpp#L37 .

If you have suggestions on better way to compute this, I'm more than open to hear & discuss about them :)

Thanks!