Re: usb1.1 vs libusb-1.0.9

[Xiaofan C. <xia...@gm...>]

On Wed, Nov 14, 2012 at 10:14 AM, Xiaofan Chen <xia...@gm...> wrote:
> On Wed, Nov 14, 2012 at 12:05 AM, geo <sha...@ya...> wrote:
>> Microcontroller with usb1.1, 32bytes buffer for bulk transfers and a lot of
>> real time data to move to Pc with Linux - kernel2.6
>> As far as I understand the maximum theoretical bandwidth available for bulk
>> transfers in this case is 19 transfers * 64 bytes / frame (1ms) = 1218
>> bytes/frame , 1218 kbytes/second
>>
>> As my bulk endpoint size is 32bytes the maximum theoretical transfer rate
>> is: 33 transfers * 32 bytes = 1056 bytes/frame, 1056 kbytes/second.
>>
>> I know this is the theoretical max , but I would be happy with around
>> 650Kbytes/second ... instead I'm getting a little less than the theoretical
>> max of 19 transfers*32bytes / frame = 608 bytes/frame ... around
>> 500bytes/frame.
>>
>> My device contains 4 bulk endpoints, 2 in , 2 out, all with 32
>> bytes simple buffer ( not double buffered ).
>
> What if you change to 64Bytes for this bulk endpoint? That may help.
> Double buffer will also help.
>
> The following is of course just for speed testing and may not be
> the real-world example. You can see it can reach 1000KB+
> speed for single direction and  500KB loopback.
> http://www.microchip.com/forums/tm.aspx?m=417509
>
> For those tests, I was using libusbdotnet with libusb-1.0
> backend under Linux. I have also carried out tests under
> Windows with either WinUSB, libusb-win32 or libusb-1.0
> backend of libusbdotnet, also using libubK's benchmark
> application.
>
> The benchmark firmware is here.
> http://code.google.com/p/usb-travis/

BTW, I have never been able to get similar performance
if I use 32bytes as the wMaxPacketSize. For example,
I only got about 300KB with the loopback test, compared
to about 560KB with 64 bytes as the wMaxPacketSize.


-- 
Xiaofan