[sane-devel] Re: [linux-usb-devel] Bad Performance uhci (vs. usb-ohci ,ohci) ->2.4.18pre9 too

[Gunther Mayer]

Johannes Erdfelt wrote:
> 
> On Tue, Feb 12, 2002, Gunther Mayer <Gunther.Mayer at t-online.de> wrote:
> > Greg KH wrote:
> > >
> > > On Sun, Feb 10, 2002 at 11:10:07AM +0100, Gunther Mayer wrote:
> > > > Hi,
> > > > the "uhci" module suffers from very bad performance (linux-2.4.17).
> > > >
> > > > Scanning 150dpi A4 (merlin670 for Canon N670U over libusb-0.1.4) gives:
> > > >   uhci:    135 sec.
> > > >   usb-uhci: 43 sec.
> > > >   ohci:     35 sec
> > >
> > > This should be fixed in the latest 2.4.18-pre patches.  Can you grab the
> > > latest one, and let us know if you still have the same problems with the
> > > uhci driver?
> >
> >
> > 2.4.18pre9 breaks "libusb-0.1.4" and so my application:
> >
> >  mc:~/libusb-0.1.4/tests # ./testlibusb
> >  bus/device  idVendor/idProduct
> >  bus=0x8049de8 dev=0x1029
> >  001
> >  Segmentation fault (core dumped)
> >
> > I added some printks, the "dev" pointer seems buggy.
> 
> You should be using 0.1.5. There are some bugs in 0.1.4

Thanks, libusb-0.1.5 is working fine with linux-2.4.18pre9 and merlin670.c

The results are about the same as linux-2.4.17 + libusb-0.1.5:

uhci:    135 sec. (the scanner has to stop and restart often)
usb-uhci: 43 sec. (the scanner stops about four times)
usb-ohci: 37 sec. (the scanner never stops! aka "streams")

To "stream", a datarate of 375KBytes/sec is needed; only usb-ohci delivers
this rate; usb-uhci is acceptable; uhci is broken performance-wise.