Real-time benchmarks
Test Bench/Reference hardware
Up-to-date reference Hardware was used. Results need to published
This is the reference hardware for the Seapath project and supported out of the box. It might cost significant time to support other hardware (e.g. AMD processors). Some integration is available with a dedicated Yocto layer: https://git.yoctoproject.org/cgit/cgit.cgi/meta-amd/
Specification
Parts | Specifications |
---|---|
Motherboard | ASMB‐823 |
Chipset | Intel C612 |
CPU | XEON 2.4G 35M 2011P 14CORE E5‐2680V4 |
Memory | 2x 8G R‐DDR4‐2400 1.2V1GX8 HYX |
Disk | SQF 2.5 SATA SSD 830 512G MLC (‐40~85°C) |
NIC | INTEL I210 NIC 10/100/1000M PCIEx4 2PORT(G) |
Tests results
Real time
Tests
With the previous test bench hardware, a couple of tests were used.
We used cyclictest:
"Cyclictest accurately and repeatedly measures the difference between a thread's intended wake-up time and the time at which it actually wakes up in order to provide statistics about the system's latencies. It can measure latencies in real-time systems caused by the hardware, the firmware, and the operating system." (source: https://wiki.linuxfoundation.org/realtime/documentation/howto/tools/cyclictest/start ).
The following arguments were provided:
cyclictest -l100000000 -m -Sp90 -i200 -h400 -q >output |
This test is very long (~5 hours).
You can then plot the latency graph:
./yocto-bsp/tools/gen_cyclic_test.sh -i output -n 28 -o output.png |
output is the output file generated by cyclictest
28 match the amount of CPU used.
output.png is the latency graph file.
we used the same arguments as used by OSADL (https://www.osadl.org/Latency-plots.latency-plots.0.html )
We created a script to plot the latency graph as done by OSADL
Results
Hypervisors
With Kernel RT Full Preempt (CONFIG_PREEMPT_RT_FULL=y)
Without Kernel RT Full Preempt (CONFIG_PREEMPT_NONE=y)
Virtual machines
All Yocto images include the ability to run guest Virtual Machines (VMs).
We used KVM and Qemu to run them. As we do not have any window manager on the host system,
VMs should be launched in console mode and their console output must be correctly set.
For testing purpose, we can run our Yocto image as a guest machine.
We do not use the .wic image which includes the Linux Kernel and the rootfs because
we need to set the console output.
We use two distinct files to modify the Linux Kernel command line:
bzImage: the Linux Kernel image
seapath-test-image-votp.ext4: the rte rootfs
Then run:
qemu-system-x86_64 -accel kvm -kernel bzImage -m 4096 -hda seapath-test-image-votp.ext4 -nographic -append 'root=/dev/sda console=ttyS0' |
Docker
You can use docker check-config.sh to check that all necessary configurations of the host linux Kernel are set: