On Wed, 16 Nov 2005, Neil Brown wrote: > I've got piles of usbmon logs, and I'm happy to create more. Could > you be specific about that would help, and different setting show > different things differently :-) > > For example, I have a rather nice log with 64 byte writes (so there is > one packet per UDB). If I take the completion time stamps, find the > remainder after dividing by 111838(microseconds) and graph that I get > a really interesting graph!! I cannot explain all the features, but > I'm sure they correlate significantly to some aspect of the printer > hardware
If it's so interesting, post it! > I guess the question is what size write request would be most helpful. > 8192bytes means there is plenty of opportunity for FSBR to push lots > of packets out, but it means that each record reported by usbmon > corresponds to 128 packets, so you lose a lot of detail. > 64bytes gives more detail, but with less chance for FSBR to have an > effect (as usblp only submits one URB at a time). > Maybe 512 would be a reasonable compromise? 64 bytes is probably best. You don't lose any FSBR performance, because FSBR isn't turned off immediately when an URB completes. > > It would also be worthwhile to see if you can get more or less the same > > effect without all those changes, just by increasing IDLE_TIMEOUT to 1000 > > ms or something like that. > > I did get similar throughput with IDLE_TIMEOUT set to 50000. 1000 > wouldn't be enough with 8K writes, as an 8K write often takes more > than one second. That's what I'd like to understand. Even with FSBR on all the time, it still takes more than 1000 ms to transfer 128 packets (i.e., 8192 bytes)? Do the packets get sent in fixed-size clumps, or is it more random? If the printer was at all sane, it wouldn't take more than 300 ms to transfer that much data even with FSBR permanently off! > But I'm happy to redo some tests to be able to give > you clear before/after results. It'll probably have to wait a week or > so, as I've spent more time on this that I really should have lately. It might be more interesting to find out how the printer manages to work more quickly under Windows (assuming it does, of course). If you want to spend the time, you could try tracing some of that communication. Alan Stern ------------------------------------------------------- This SF.Net email is sponsored by the JBoss Inc. Get Certified Today Register for a JBoss Training Course. Free Certification Exam for All Training Attendees Through End of 2005. For more info visit: http://ads.osdn.com/?ad_id=7628&alloc_id=16845&op=click _______________________________________________ linux-usb-devel@lists.sourceforge.net To unsubscribe, use the last form field at: https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
