Thanks David
David Brownell wrote: > > I deduce that you're not using urb queueing. If you keep an I/O queue > full at all times, you'll be limited by hardware performance of your > FX2 and EHCI controllers ... I tried using the BULK_QUEUE flag and then simply submiting URBs, but the copy_to_user did not work then (as I remember - I have to check again tonight). I have to submit the next URB only after the bulk_irq was received. Are there special functions I need to use for URB queueing? Or are you saying, just simply submit all the URBs and let the Linux USB subsystem handle the queueing? Are URB memory sizes important? Should they remain constant? > You'll probably need to have separate activities reading data and > writing it to disk. The writing is done in user-space. I think I may make the write in a thread. Is this what you are saying? > I've had one person report 39 MB/sec on Linux, > FX2-to-disk, that way ... I think that was with a VT8235. Awesome, I hope I can match that, this is an Intel chip that is on an Intel motherboard. > I'd hope that any new usermode USB APIs will all be designed around > new kernel mode ones ... using the 2.6 kernel and libaio. I am not familiar with libaio and I am developing on 2.4.22 (whatever the latest stable version is that comes with MDK 9.2). I am simply using devfs with file_operations, and character device read(). > > Is this a hardware problem that we do not get 60MB/s (or 52MB/s)? > > Hopefully, this will be solved in the future. > > The FX2 doesn't have much of a CPU. The max bulk throughput > is 52 MB/s, the 60 MB number doesn't count framing and other > limitations. And I'm not sure I'd expect any hardware to > really hit the 52 MB/s number. I would like to see bit-stuffing dropped for USB 3.0, bit-stuffing is not needed for ieee1394, the CRC should be enough. Max out the USB speed!!! > > I am going to test bulk transfers on XP and I will report the results. > > According to one Cypress document the FX2 averages 39MB/s in XP (so > > 9MB/s faster than I am getting on Linux, but still under the 50MB/s > > claim). Here are my fx2 results in XP: BULK IN = 34.5 MB/s [doc has 39.967] CPU=16% BULK OUT=31.7MB/s [31.815] CPU=16% ISO IN=23.976MB/s [24.000] CPU=5% ISO OUT=23.975MB/s [23.975] CPU=5% I am surprised that the BULK IN did not compare with the FX2 document. I was hoping for that 40MB/s, and that does not even including writing (although there is enough CPU ). thanks for your advice on this! Ted -- Ted Huntington Programmer Analyst I Main Library University of California, Irvine PO Box 19557 Irvine, CA 92623-9557 emesgs: [EMAIL PROTECTED] web page: http://business.lib.uci.edu/webpages/ted.htm 8:00a-12:00p Business Office (949) 824-8926 1:00p-5:00p Multimedia Resource Center (949) 824-1674 "Stop violence, teach science." ------------------------------------------------------- SF.Net is sponsored by: Speed Start Your Linux Apps Now. Build and deploy apps & Web services for Linux with a free DVD software kit from IBM. Click Now! http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click _______________________________________________ [EMAIL PROTECTED] To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
