quick2wire GPIO is much slower than other python modules

[RandomDude4]

It takes around 0.5 ms to change the output of one pin with the quick2wire module for me.

The "python-rpi.gpio" module found on apt-get, does the same operation in around 0.01 ms (more than 10x faster)

Although both of these are very slow, a 700 MHz processor would be able to do it in a few nanoseconds, but a slowdown of 100000 times sounds very inefficient even for python?

[npryce]

Are you running as root or as an unpriviledged user?

[RandomDude4]

It does not matter if I'm root or not, however if I use the following (value change is done in a loop with timing before and after):

outpin = Pin(11, "out")
outpin.value = 1

I get around 0.4-0.5 ms per change, but with the code:

Pin(11, "out")
Pin(11).value = 1

I get even slower speeds of 2 ms per output change.

I have the quick2wire-python-api "installed" through PYTHONPATH=/xxx/quick2wire-python-api/src.

Edit:
Running in python 2.7 is faster than python 3.2 (I thought it was written for python 3, didn't even know it worked in python 2.7 untill now)
#!/usr/bin/env python --> 0.1 ms per GPIO change
#!/usr/bin/env python3 --> 0.45 ms per GPIO change

Although still 10x slower than "python-rpi.gpio" module = 0.015 ms per GPIO change in both python 3.2 and 2.7 (same script, only changed modules and syntax)

[npryce]

Sorry, I didn't make myself clear. I meant, is running as an unprivileged user important to you?

RPi.GPIO maps the GPIO device registers into the memory space of the process, which can only be done as root. This increases the attack surface you're driving GPIO from a network service exposed on the Internet. It's also a hassle during development because programming errors can crash the machine or trash your filesystem.

We're designing our API for convenience and security before raw performance. Our API goes through the user-space GPIO API, writing to files in the /sys filesystem. The kernel transition and driver layer takes a lot of time compared to direct writes to the device registers. However, our API can then be used by unprivileged processes that cannot bring down the entire machine.

It's a trade-off. If you're ok about running as root and need raw speed, say in a robotics project or if bit-banging a binary protocol, then RPi.GPIO is a better choice. If you're connecting sensors to the Internet and raw speed is not critical then our API would be a better choice (in my opinion).

We're working on closing the performance gap, safely mapping the device registers into user space without having to run as root. But that's work in progress.

[npryce]

Interesting that Python2 and Python3 have such different timings. I'll look into that. I suspect that maybe there's some unnecessary unicode to binary conversion being performed in Python3

[RandomDude4]

Now I understand, in that case I am happy with the speed from your API, it's definitely more trouble having to run everything as root. I didn't know the difference between RPi.GPIO and your API before, I thought booth went through the kernel device drivers in the same manner.

Feel free to remove this issue if you want to.

[npryce]

Closed as duplicate of #6, new issue #19 raised for python 2 / python 3 performance discrepancy.