Re: usb1.1 vs libusb-1.0.9

[Vincent P. <plr...@gm...>]

Le 14 nov. 2012 19:07, "geo" <sha...@ya...> a écrit :
> Yes, in the rx callback ( cb_xfr ) I submit each time a new transfer,
8/16Kbytes in size.

You may prefer to resubmit the existing transfer rather than creating a new
one. Not that I expect it to be a bottleneck on fullspeed usb on any decent
host compute.

Note also that (as already mentioned in the thread), you really want to
have more than one transfer pending at all time, for a symetric reason that
double buffering was advised on uC side: while host is busy handling
incomming data, kernel has a buffer to put incomming data to.

> 2 bulk endpoints + 1interrupt are used for a CDC console

IIRC (no doc at hand) interrupt transfers steal bandwidth, at least in
device-to-host direction: host polls device actually (usb transaction
always being initiated by host). I cannot tell the impact though, just
raising the point.

Regards,
Vincent Pelletier