On Tue, 13 Mar 2001, Evgenii Stoev wrote:

> We develop usb device with bulk transfer at full speed . We have speed
> problem. The time for one transaction ( IN or OUT) is aprox. 10 - 25 ms.

well, first to catch the more obvious, it depends on the length of your
transaction of course. If there are several 10kB to transfer it will take
some 10ms for sure. Furthermore this depends on the remaining bandwidth
for bulk transactions (might be 0 if all allocated to control/int/iso
traffic) and how fast the device reacts to the bulk EP request.

> We use the function usb_bulk_msg. Any ideas how to reduce this time. I
> send my write and read functions.

Since you are using the blocking api you are probably limited by the
scheduling latency. You might want to use the nonblocking api and
measure the delay between calling submit_urb() and your completion handler
gets called.
Under optimal conditions I've measured latencies and throughput really
close to the optimum: about 1ms latency per kB transfer length and up to
1.1MB/s continous average throughput for usb-ohci (0.9MB/s for  [usb-]uhci)
from bulk-in EP. The device probably has to provide double buffering to
achieve these values.

Martin


_______________________________________________
[EMAIL PROTECTED]
To unsubscribe, use the last form field at:
http://lists.sourceforge.net/lists/listinfo/linux-usb-devel

Reply via email to