2011/5/9 Emmanuel Blot <eblot...@gmail.com>: > Hello, > > We have been facing performance issues with PyUSB. > > As our code can run with both pyftdi/libftdi/libusb (Py/C/C) or > pyftdi/pyusb/libusb (Py/Py/C), we have profiled the code and observed that > the stack based on the native libftdi librarie performs far better than the > one based on the pyusb stack. It was quite surprising to see a huge > difference between both stack, as libftdi really is a thin wrapper on top of > libusb. > > With the help of the Python profiling modules, we have been able to find the > culprit: a sub-optimized call to the read() and write() functions. These > functions accept an 'interface' parameter, which can be of different type - > such as an interface number, None or an core.Interface class instance. > > When this parameter is not an core.Interface object, the read() and write() > methods invoke -through the get_interface() method - a time-costly interface > resolution processing. This routine is - relative to the other calls - deadly > slow, and is repeated for every single read() or write() call. > > Replacing the interface index (an integer) with a core.Interface object shows > a tremendous improvement of the average transmission time, with final > performances very close to the native C implementation of the libftdi library. > > I'm not sure this hint is documented somewhere, but I thing it could be part > of the FaQ or something, with an reference from the Python documentation > string, as selecting the proper parameter type for the read() and write() > 'interface' shows a large performance boost. > The root of all trouble is the internal _ResourceManager.get_configuration method, that returns an interface object. If I change it to return the interface number instead, maybe the performance gets improved, either with an Interface object or interface number as parameter. Going to check that.
Wander ------------------------------------------------------------------------------ Achieve unprecedented app performance and reliability What every C/C++ and Fortran developer should know. Learn how Intel has extended the reach of its next-generation tools to help boost performance applications - inlcuding clusters. http://p.sf.net/sfu/intel-dev2devmay _______________________________________________ pyusb-users mailing list pyusb-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/pyusb-users
