|
From: Michael P. <mic...@gm...> - 2009-07-24 15:15:00
|
2: A) With a totally different project (i.e., not libusb), I've found that my code can behave funny over an ssh connection if I fail to make stdout nonblocking (fcntl(fileno(stdout), F_SETFL, oldflags | O_NONBLOCK)) when the ssh connection is likely to drop out. B) Locally, on older computers, I've noticed that certain very long displays which I don't care about can be sped up by temporarily switching to a different virtual console. Both of these indicate you may be slowed down in some cases. You might just try it and print your own (condensed) error info instead, if possible. Does the display appear to lag? 1: I don't know, but the frames are scheduled every 1 ms. I've heard something about doing async transfers, and queueing multiple transfers up to increase the likelihood you're scheduled on EVERY 1ms frame. You can theoretically schedule quite a few 64-byte packets in that 1ms (though not quite 12Mbps). I've never tried it, but you might look there. The bulk transfers are not guaranteed BW afaik, so they might suffer from high traffic. Michael -----Original Message----- From: Rick Lamanna [mailto:Ric...@mc...] Sent: Friday, July 24, 2009 9:55 AM To: lib...@li... Subject: [Libusb-devel] Bulk transfer perfromance Hello, I have a couple of questions on bulk transfers. 1. I have a device that scans data at various clock speeds. Unfortunately, the device has a very small buffer. At the device's maximum rate, the buffer fills up in 2.5 mS. I read data from the buffer using bulk transfers. The device has a max packet size of 64 bytes. Can I expect to read data from the device's buffer fast enough using Libusb so that I don't lose data? (actually the device will go into an overrun state if I don't read fast enough) 2. Does setting the Libusb debug level to 3 impact the performance of doing bulk transfers? I testing the throughput of my device - should I set the debug level to 0? Thanks Rick L |