Hi Victor, > On Sep 23, 2015, at 17:27 PM, Victor Toso <li...@victortoso.com> wrote: > > Hey, > > I was taking a look at rhbz#1264156 [1] which leads to a great amount > of memory being used in the client when sharing webcam with usbredir > when the connection is slow enough (not necessary slow but slower then > the webcam data flow). > > Looks like there are other bugs about similar issue like rhbz#1201387 > [2] with _being slow_ and high memory consumption in the client. I've > also hit fdo#90286 [3] during my local tests... > > [1] https://bugzilla.redhat.com/show_bug.cgi?id=1264156 > [2] https://bugzilla.redhat.com/show_bug.cgi?id=1201387 > <https://bugzilla.redhat.com/show_bug.cgi?id=1201387> > [3] https://bugs.freedesktop.org/show_bug.cgi?id=90286 > <https://bugs.freedesktop.org/show_bug.cgi?id=90286> > Well, what I was planing to do in order to address the high memory > consumption is the following combination: > > 1-) Avoid sharing devices when connection is too slow. > libusb has libusb_get_device_speed function which can possibility > return the operation speed of the device. By having the quality and > speed of the connection we could deny connecting devices that are > too fast. > > 2-) Limiting the queue size for usbredir > the memory grows because the queue to send the data grows faster > then network's read. Even with (1) we could hit issues in the long > run. In this case, we could identify by the queue limit being reached. > The solution should be disabling the usbredirection of the device.
This solution is the most robust one probably. > > 3-) Negotiate capabilities with device > Not sure how much we can configure from spice-gtk to device > parameters but theoretically it should be possible to change resolution > and quality of the device, making less use of bandwidth. This is an interesting idea, but will require device/class-specific code unfortunately. It would be nice to use Bandwidth Allocation mechanisms somehow, but I can’t think of a technically acceptable way… ~Dmitry. > > That's the general idea to address those issues and after talking with > Pavel we agreed that would be good to send an email before diving into > it. > > Any thoughts about this? Hints? Patches? :) > > side-note: > I'm fairly convinced that channel-usbredir could trigger memory > issues in the server side as happened with channel-webdav in [4] and > [5] although I wasn't able to trigger it yet. > > [4] https://bugs.freedesktop.org/show_bug.cgi?id=91350 > [5] https://bugs.freedesktop.org/show_bug.cgi?id=91903 > > cheers, > Victor Toso > _______________________________________________ > Spice-devel mailing list > Spice-devel@lists.freedesktop.org > http://lists.freedesktop.org/mailman/listinfo/spice-devel
_______________________________________________ Spice-devel mailing list Spice-devel@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/spice-devel
