Ethernet Device Performance

Network Layout

If you face a lots of lost packets, the first thing to try is to directly connect the device to your machine, using a wired connection. If it is possible, use a dedicated port.

Most of the cameras are able to saturate their gigabit link, so any additional traffic may lead to packet lost. If your machine ethernet port has a lower bandwidth than the camera, make sure the stream you are requesting will not exceed its bandwidth.

Socket Buffer Size

Under heavy CPU load, the input socket buffer size may be increased in order to avoid lost packets because of a full buffer. You can try the --auto parameter of arv-camera-test.

In you application, this can be tweaked using g_object_set() on the stream object. For example, this code, to be called before arv_camera_start_acquisition(), will enable the auto buffer size mode:

g_object_set (stream,
          "socket-buffer", ARV_GV_STREAM_SOCKET_BUFFER_AUTO,
          "socket-buffer-size", 0,
          NULL);

Receiving Thread Priority

It is possible to increase the receiving thread priority. You can experiment with thread priority using the --realtime or --high-priority options of arv-camera-test.

In your code, stream thread priority can be changed in the stream buffer callback:

static void
stream_cb (void *user_data, ArvStreamCallbackType type, ArvBuffer *buffer)
{
        if (type == ARV_STREAM_CALLBACK_TYPE_INIT) {
        if (!arv_make_thread_realtime (10))
            printf ("Failed to make stream thread realtime\n");
    }
}

or:

static void
stream_cb (void *user_data, ArvStreamCallbackType type, ArvBuffer *buffer)
{
        if (type == ARV_STREAM_CALLBACK_TYPE_INIT) {
        if (!arv_make_thread_high_priority (-10))
            printf ("Failed to make stream thread high priority\n");
    }
}

Stream Packet Size

One way to increase streaming performance and lower the CPU use is to increase the stream packet size. arv_camera_gv_set_packet_size() and arv_camera_gv_auto_packet_size() allow you to change this parameter. By default, the network adapter of your machine will probably not let you receive packet bigger than 1500 bytes, which is the default Maximum Transfer Unit (MTU). It means if you want to use big packets, you also have to increase the network adapter MTU to a greater walue (8192 bytes being a recommended value). The exact procedure to change the MTU depends on the distribution you are using. Please refer to the administrator manual of your distribution.

On Fedora, MTU can be changed using the network setting panel. Edit the wired network settings. MTU parameter is in the Identity page.

On Debian / Ubuntu you can add a line “mtu 8192” to /etc/network/interfaces under the correct device.

iface ethusb0 inet static
    address 192.168.45.1
    netmask 255.255.255.0
    broadcast 192.168.45.255
    mtu 8192

Please note if your device is not connected directly to the machine, you may also have to tweak the active devices on your network.

Packet Socket Support

Aravis can use packet sockets for the video receiving thread. But this mode requires extended capabilities. If you want to allow your application to use packet socket, you must set the cap_net_raw capability using setcap. For example, the following command gives this capability to the Aravis viewer:

sudo setcap cap_net_raw+ep arv-viewer

Legacy endianess mechanism

Some GigEVision devices incorrectly report a Genicam schema version greater or equal to 1.1, while implementing the legacy behavior for register access. We maintain a list in arvgcport.c which allows to force the use of the legacy endianness mechanism.

The documentation about the legacy endianness mechanism is in the 3.1 appendix (‘Endianess of GigE Vision Cameras’) of the GenICam Standard.

There is a chance this part of Aravis is due to a misunderstanding of how a GigEVision device is supposed to behave (Remember we can not use the GigEVision specification documentation). But until now, there was no evidence in the issue reports it is the case. If you think this should be implemented differently, don’t hesitate to explain your thoughts on Aravis issue report system, or even better, to open a pull request. The related Aravis issues are available here: https://github.com/AravisProject/aravis/labels/5. Genicam 1.0 legacy mode.

Meanwhile, if you want to add a device to the exception list, please open an issue on github, giving the vendor name and model name as found in the Genicam data of your device, using the genicam command of arv-tool. They are stored in the ModelName and VendorName attributes of the RegisterDescription element.