SuperDARN Canada uses OpenSUSE for an operating system, but any Linux system that can support the NVIDIA drivers for the graphics card will work. The current latest version of OpenSuSe (15.3) is known to work. Commands that require root privileges will have a `sudo` or `su` command ahead of them, or explicitly say ‘as root’, all others should be executed as the normal user (recommended name: radar) that will run Borealis
Install the latest version of the NVIDIA drivers (see https://en.opensuse.org/SDB:NVIDIA_drivers). The driver must be able to support running the GPU selected and must also be compatible with the version of CUDA that supports the compute capability version of the GPU. Getting the OS to run stable with NVIDIA is the most important step, so make sure you read this page carefully.
Use the BIOS to find a stable over-clock for the CPU. Usually the recommended turbo frequency is a good place to start. This step is optional, but will help system performance when it comes to streaming high rates from the USRP. Do not adjust higher over-clock settings without doing research.
Use the BIOS to enable boot-on-power. The computer should come back online when power is restored after an outage. This setting is typically referred to as Restore on AC/Power Loss
Use cpupower to ungovern the CPU and run at the max frequency. This should be added to a script that occurs on reboot.
sudo cpupower frequency-set -g performance.
To verify that the CPU is running at maximum frequency:
Use ethtool to set the interface ring buffer size for both rx and tx. Make sure to use an ethernet device which is connected to the 10 Gb card of the computer (not necessarily eth0). This should be added to a script that occurs on reboot for the interface used to connect to the USRPs. This is done to help prevent packet loss when the network traffic exceeds the capacity of the network adapter.
sudo ethtool -G <10G_network_device> tx 4096 rx 4096.
To see that this works as intended, and that it persists across reboots, you can execute the following, which will output the maximums and the current settings.
sudo ethtool -g <10G_network_device>
Use the network manager or a line in the reboot script to change the MTU of the interface for the interface used to connect to the USRPs. A larger MTU will reduce the amount of network overhead. An MTU larger than 1500 bytes allows what is known as Jumbo frames, which can use up to 9000 bytes of payload.
sudo ip link set <10G_network_device> mtu 9000
To verify that the MTU was set correctly:
ip link show <10G_network_device>
Use sysctl to adjust the kernel network buffer sizes. This should be added to a script that occurs on reboot for the interface used to connect to the USRPs. That’s 50 million for rmem_max and 2.5 million for wmwem_max.
sudo sysctl -w net.core.rmem_max=50000000
sudo sysctl -w net.core.wmem_max=2500000
Verify that the kernel network buffer sizes are set:
The previous commands should all be executed on boot, as they are not persistent. One way to do this is via the root user’s crontab. Here are some example entries for a computer with multiple ethernet interfaces:
@reboot /sbin/sysctl -w net.ipv6.conf.all.disable_ipv6=1
@reboot /sbin/sysctl -w net.ipv6.conf.default.disable_ipv6=1
@reboot /usr/sbin/ethtool -G <10G_network_device_1> tx 4096 rx 4096
@reboot /usr/sbin/ethtool -G <10G_network_device_2> tx 4096 rx 4096
@reboot /sbin/ip link set dev <10G_network_device_1> mtu 9000
@reboot /sbin/ip link set dev <10G_network_device_2> mtu 9000
@reboot /usr/bin/cpupower frequency-set -g performance
@reboot /sbin/sysctl -w net.core.rmem_max=50000000
@reboot /sbin/sysctl -w net.core.wmem_max=2500000
Install tuned. Use tuned-adm (as root) to set the system’s performance to network-latency.
sudo zypper in tuned
systemctl enable tuned
systemctl start tuned
tuned-adm profile network-latency
To verify the system’s new profile:
sudo tuned-adm profile_info
Add an environment variable called BOREALISPATH that points to the cloned git repository in .bashrc or .profile and re-source the file. For example:
Verify the BOREALISPATH environment variable exists:
env | grep BOREALISPATH
Clone the Borealis software to a directory The following ensures that Borealis will be in the same directory that the `BOREALISPATH` env variable points to.
sudo zypper in git
git clone https://github.com/SuperDARNCanada/borealis.git $BOREALISPATH
If you’re building Borealis for a University of Saskatchewan radar, complete the following steps. If not, skip ahead to the next step. Create symlink config.ini in borealis directory and link to the site specific config file.
git submodule init && git submodule update
ln -svi $BOREALISPATH/borealis_config_files/[radarcode]_config.ini config.ini
If you’re building Borealis for a non University of Saskatchewan radar, use a Usask config.ini file (located here) as a template or the config file documentation to create your own file in the borealis directory.
In config.ini, there is an entry called “realtime_address”. This defines the protocol, interface, and port that the realtime module uses for socket communication. This should be set to “realtime_address” : “tcp://<interface>:9696”, where <interface> is a configured interface on your computer such as “eth0” or “wlan0”. Running ip addr, you should choose a device which is UP.
The Borealis software has a script called install_radar_deps.py to help install dependencies. This script has to be run with root privileges. This script can be modified to add the package manager of a different distribution if it doesn’t exist yet. Make sure that the version of CUDA is up to date and supports your card. This script makes an attempt to correctly install Boost and create symbolic links to the Boost libraries the UHD (USRP Hardware Driver) understands. If UHD does not configure correctly, an improper Boost installation or library naming convention is the likely reason. Note that you need python3 installed before you can run this script. The radar abbreviation should be the 3 letter radar code such as ‘sas’, ‘rkn’ or ‘inv’.
chmod +x install_radar_deps.py
python3 install_radar_deps.py [radar abbreviation] $BOREALISPATH > install_log.txt 2>&1
Install the necessary software to convert and move/copy data:
Set up NTP. The install_radar_deps.py script already downloads and configures a version of ntpd that works with incoming PPS signals on the serial port DCD line. An example configuration of ntp is shown below for /etc/ntp.conf. These settings use tick.usask.ca as a time server, and PPS (via the 127.127.22.0 lines). It also sets up logging daily for all stats types.
driftfile /var/log/ntp/ntp.drift statsdir /var/log/ntp/ntpstats/ logfile /var/log/ntp/ntp_log logconfig =all statistics loopstats peerstats clockstats cryptostats protostats rawstats sysstats filegen loopstats file loopstats type day enable filegen peerstats file peerstats type day enable filegen clockstats file clockstats type day enable filegen cryptostats file cryptostats type day enable filegen protostats file protostats type day enable filegen rawstats file rawstats type day enable filegen sysstats file sysstats type day enable restrict -4 default kod notrap nomodify nopeer noquery limited restrict -6 default kod notrap nomodify nopeer noquery limited restrict 127.0.0.1 restrict ::1 restrict source notrap nomodify noquery server tick.usask.ca prefer server 127.127.22.0 minpoll 4 maxpoll 4 fudge 127.127.22.0 time1 0.2 flag2 1 flag3 0 flag4 1 keys /etc/ntp.keys trustedkey 1 requestkey 1 controlkey 1
Edit /etc/security/limits.conf (as root) to add the following line that allows UHD to set thread priority. UHD automatically tries to boost its thread scheduling priority, so it will fail if the user executing UHD doesn’t have permission.
@users - rtprio 99
Assuming all dependencies are resolved, use scons to build the system. Use the script called mode to change the build environment to debug or release depending on what version of the system should be run. SCONSFLAGS variable can be added to .bashrc/.profile to hold any flags such as -j for parallel builds. For example, run the following:
source mode [release|debug]
If first time building, run scons -c to reset project state.
scons to build
Add the Python scheduling script, start_radar.sh, to the system boot scripts to allow the radar to follow the schedule. As an example on OpenSuSe for the radar user:
Add the line @reboot /home/radar/borealis/start_radar.sh >> /home/radar/start_radar.log 2>&1
Create necessary directories. Here is an example for a user named radar and the standard configuration in the ‘config.ini’ file:
sudo mkdir -p /data/borealis_logs
sudo mkdir -p /data/borealis_data
sudo chown radar:users /data/borealis_logs
sudo chown radar:users /data/borealis_data
Find out which tty device is physically connected to your PPS signal. It may not be ttyS0, especially if you have a PCIe expansion card. It may be ttyS1, ttyS2, ttyS3 or higher. To do this, search the system log for ‘tty’ (either dmesg or the syslog). An example output with a PCIe expansion card is below. The output shows the first two ttyS0 and 1 are builtin to the motherboard chipset and are not accessible on this x299 PRO from MSI. The next two ttyS4 and S5 are located on the XR17V35X chip which is located on the rosewill card:
[ 1.624103] serial8250: ttyS0 at I/O 0x3f8 (irq = 4, base_baud = 115200) is a 16550A [ 1.644875] serial8250: ttyS1 at I/O 0x2f8 (irq = 3, base_baud = 115200) is a 16550A [ 1.645850] 0000:b4:00.0: ttyS4 at MMIO 0xfbd00000 (irq = 37, base_baud = 7812500) is a XR17V35X [ 1.645964] 0000:b4:00.0: ttyS5 at MMIO 0xfbd00400 (irq = 37, base_baud = 7812500) is a XR17V35X
Try attaching the ttySx line to a PPS line discipline using ldattach:
/usr/sbin/ldattach PPS /dev/ttyS[0,1,2,3,etc]
Verify that the PPS signal incoming on the DCD line of ttyS0 (or ttySx where x can be any digit 0,1,2,3…) is properly routed and being received. You’ll get two lines every second corresponding to an ‘assert’ and a ‘clear’ on the PPS line along with the time in seconds since the epoch. If it’s the incorrect one, you’ll only see a timeout.
sudo ppstest /dev/pps0 [sudo] password for root: trying PPS source "/dev/pps0" found PPS source "/dev/pps0" ok, found 1 source(s), now start fetching data... source 0 - assert 1585755247.999730143, sequence: 200 - clear 1585755247.199734241, sequence: 249187 source 0 - assert 1585755247.999730143, sequence: 200 - clear 1585755248.199734605, sequence: 249188
If you’re having trouble finding out which /dev/ppsx device to use, try grepping the output of dmesg to find out. Here’s an example that shows how pps0 and 1 are connected to ptp1 and 2, pps2 is connected to /dev/ttyS0 and pps3 is connected to /dev/ttyS5.:
[ 0.573439] pps_core: LinuxPPS API ver. 1 registered [ 0.573439] pps_core: Software ver. 5.3.6 - Copyright 2005-2007 Rodolfo Giometti <firstname.lastname@example.org> [ 8.792473] pps pps0: new PPS source ptp1 [ 9.040732] pps pps1: new PPS source ptp2 [ 10.044514] pps_ldisc: PPS line discipline registered [ 10.045957] pps pps2: new PPS source serial0 [ 10.045960] pps pps2: source "/dev/ttyS0" added [ 227.629896] pps pps3: new PPS source serial5 [ 227.629899] pps pps3: source "/dev/ttyS5" added
Now add the GPS disciplined NTP lines to the root startup script using the tty you have your PPS connected to.
/sbin/modprobe pps_ldisc && /usr/sbin/ldattach PPS /dev/[PPS tty] && /usr/local/bin/ntpd
Verify that the realtime module is able to communicate with other modules. This can be done by running the following command in a new terminal while borealis is running. If all is well, the command should output that there is a device listening on the channel specified.
ss –all | grep 9696
For further reading on networking and tuning with the USRP devices, see https://files.ettus.com/manual/page_transport.html and https://kb.ettus.com/USRP_Host_Performance_Tuning_Tips_and_Tricks. Also see http://doc.ntp.org/current-stable/drivers/driver22.html for information about the PPS ntp clock discipline, and the man pages for: