On Thu, Sep 10, 2009 at 08:16:31PM +0300, Aleksandr V. Piskunov wrote: > On Thu, Sep 10, 2009 at 05:48:22PM +0300, Antti Palosaari wrote: > > On 09/10/2009 04:47 PM, Antti Palosaari wrote: > >> Aleksandr V. Piskunov wrote: > >>> On Thu, Sep 10, 2009 at 04:12:15PM +0300, Antti Palosaari wrote: > >>>> Aleksandr V. Piskunov wrote: > >>>>>> Here is a test case: > >>>>>> Two DVB-T USB adapters, dvb_usb_af9015 and dvb_usb_af9015. > >>>>>> Different tuners, > >>>>> Err, make it: dvb_usb_af9015 and dvb_usb_ce6230 > >>>> Those both uses currently too small bulk urbs, only 512 bytes. I have > >>>> asked suitable bulk urb size for ~20mbit/sec usb2.0 stream, but > >>>> no-one have answered yet (search ml back week or two). I think will > >>>> increase those to the 8k to reduce load. > >>>> > >>> > >>> Nice, I'm ready to test if such change helps. > >> > >> OK, I will make test version in couple of hours. > > > > Here it is, USB2.0 URB is now about 16k both af9015 and ce6230 devices. > > Now powertop shows only about 220 wakeups on my computer for the both > > sticks. > > Please test and tell what powertop says: > > http://linuxtv.org/hg/~anttip/urb_size/ > > > > I wonder if we can decide what URB size DVB USB drivers should follow > > and even add new module param for overriding driver default. > > Thanks, Antti! > > Tested your branch on affected system. > > Load definitely went down, from ~7000 wakeups to ~250 for each tuner > according to powertop. > Both tuners still working ok if not used simultaneously or if used the > same time on different USB controllers. > > Bad news are that original problem still persists: putting both tuners > on same USB controller and zapping simultaneously corrupts stream. > Interesting observation: no matter in what sequence tuners are connected > (i.e. become adapter0 or adapter1), af9015 stream always gets heavily > distorted, visually mplayer picture becomes like 80% corrupted with > random color blocks and pixels, sound becomes a mess. At the same time > ce6230 gets slight corruption, a few discolored blocks at the time and > sound hickups. > > Anyway, will try to do a few more tests: > 1) Two usb flash drives on same controller calculating md5sum of > big .iso file, to check if it is/isn't dvb-usb problem. > 2) Will see if same issue persists on another PC with same motherboard > (slightly different revision) to rule out hardware issues. If I manage > to wire antenna there, that is... Ok, two USB flash drives on same controller, no problem when bulk reading from both at the same time, no speed drops, no corruption. Now if I plug ce6230 tuner, zap to channel and then start reading from flash drive: * slightly corrupted TS stream * flash drive read getting starved on bandwidth, speed drops from 10 MB/s to ~7 MB/s If I plug af9015 tuner, zap and read from flash * heavy corruption of TS stream * flash drive read speed drops from 10 MB/s to 2(!) MB/s Now I don't really know the USB protocol under-the-hood details, all the different types of bandwidth, reservation and so on. But shouldn't one 480 Mbit/sec controller handle rather large number of digital tuners, each pushing 20-25 Mbit/sec max, even considering all the overhead? -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
