BANANAPI.md

# Setup BananaPI for benchmarking

The goal of this documentation is to get a linux image running on a
bananaPI board that allows for very isolated benchmarks showing full
time distributions of the measurement runs.

## Base Setup

First step is to get a linux image running on the banana PI. We choose to use the
[armbian](https://www.armbian.com/) project as it is the only one with support for the
bananaPI m3. For this we generally 
[follow the instructions given](https://docs.armbian.com/Developer-Guide_Build-Preparation/),
below are notes on what to do to get the rt kernel patch into it and to build.

You can also use our pre-build image [on google drive](https://drive.google.com/open?id=1RiHymBO_XjOk5tMAL31iOSJGfncrWFQh)
and skip the build process below. Just use etcher (https://www.balena.io/etcher/) or similar,
flash an sd card and the PI should boot up. Default login is root/1234, follow the instructions,
then continue with the isolating system setup steps for more accurate measurements.

General Setup:
- Setup an ubuntu bionic 18.04 virtual box VM
- `# apt-get -y -qq install git`
- `$ git clone --depth 1 https://github.com/armbian/build`
- `$ cd build`
- To verify the environment first go for a 'clean build' without patch: `# ./compile.sh` 
- Select the bananaPI m3 board and a minimal console build

Apply RT Pach:
- Find the current kernel version armbian is using, e.g. from the previous build logs
- Download and unpack the matching rt path from https://mirrors.edge.kernel.org/pub/linux/kernel/projects/rt/
- You should have a single .patch file, place it in build/userpatches/kernel/sunix-current/patch-5.4.28-rt19.patch
- Re-run the `# ./compile.sh` script
- Select BananaPI M3, Command Line Minimal and SHOW KERNEL CONFIG
- The build should pick up the patch (and show it in the logs)
- You will be ask to fill in some settings. Choose (4) fully preemptive at the first option
- Fill out the other asked settings to your liking. To avoid issues just leave them at default.
- You will then be in the kernel config window
- Here disable the file systems AUFS and NFS in the settings (they cause build issues and we do not need them)
- Store the settings and build the image
- If successfull, the flashed image should show the preempt patch with `uname -a` and should have good latencies in cyclictest

## Run project

First setup some base dependencies for running the benchmark and tests:

- `# apt-get install rt-tests`
- `# apt-get install build-essential`
- `# apt-get install cmake`
- `# apt-get install git`
- `# apt-get install cpuset`

Next EMBB is required as a comparison in the benchmark suite. Install it using the following or similar
(as described on their github page, https://github.com/siemens/embb):
- `$ wget https://github.com/siemens/embb/archive/v1.0.0.zip`
- `$ unzip v1.0.0.zip`
- `$ cd embb-1.0.0`
- `$ mkdir cmake-build-release`
- `$ cd cmake-build-release`
- `$ cmake ../`
- `$ cmake --build .`
- `# cmake --build . --target install`

This are all dependencies needed for executing the benchmark project and pls itself.
Follow the project specific instructions for how to use them.

## Tweaking Scheduler, CPU and Interrupts

We would like to get very little dispersion through system jitter. We recommend tweaking the
scheduler, CPU and interrupt settings before running benchmarks.

See the sub-sections below for the individual measures. ***Before running tests make sure to
run the following scripts:***
- `sudo ./setup_cpu.sh`
- `sudo ./map_interrupts_core_0.sh`
- `sudo ./setup_rt.sh`
- `sudo ./setup_cgroups.sh`

Then start your tests manually mapped to cores 1 to 7. We also found that having any interactive sessions
open during the measurements (especially)

### Tuning kernel parameters

Some online references advice on some kernel parameter tweaks for getting better latencies.
To change kernel parameters edit the `boot/armbianEnv.txt` file and add a line with
`extraargs=<your args>`.

Here are some good articles discussing jitter on linux systems:
- https://www.codethink.co.uk/articles/2018/configuring-linux-to-stabilise-latency/ (General Tips and Measurements)
- https://access.redhat.com/sites/default/files/attachments/201501-perf-brief-low-latency-tuning-rhel7-v1.1.pdf  (7 - Kernel Command Line)
- https://access.redhat.com/articles/65410 (Power Management/C-States)
- https://community.mellanox.com/s/article/rivermax-linux-performance-tuning-guide--1-x (General Tips)

We use the following settings: 
```shell script
mce=ignore_ce nosoftlockup nmi_watchdog=0 transparent_hugepage=never processor.max_cstate=1 idle=poll nohz=on nohz_full=1-7
```

- ***mce=ignore_ce*** do not scan for hw errors. Reduce the jitter introduced by periodic runs
- ***nosoftlockup*** do not log backtraces for tasks hogging the cpu over some time. This, again, reduces jitter and we do not need the function in our controlled test environment.
- ***nmi_watchdog=0*** disables the nmi watchdog on architectures that support it. Esentially disables a non blockable interrup that is used to detect hanging/stuck systems. We do not need this check during our benchmarks. https://medium.com/@yildirimabdrhm/nmi-watchdog-on-linux-ae3b4c86e8d8
- ***transparent_hugepage=never*** do not scan for small pages to combine to hugepages. We have no issues with memory usage, spare us of this periodic jitter. 
- ***processor.max_cstate=1 idle=poll*** do not switch to CPU power saving modes (c-states). Just run all cores at full speed all the time (we do not care about energy during our tests).
- ***nohz=on nohz_full=1-7*** disable houskeeping os ticks on our isolated benchmark cores. core 0 will handle these when needed.

### Pin all other processes to core 0 (crgoups)

We want to isolate our measurements to cores 1 to 7 and use core 0 for all non benchmark related processes.
isolcpus is often used for this, however, we found that it disables the scheduler from balancing tasks
between the isolated cores. A better approach is to use cgroups. 
See the tutorial for further information: https://github.com/lpechacek/cpuset/blob/master/doc/tutorial.txt
Essentially, we can partition our cores into two isolated groups, then map all tasks that can be moved away from
our benchmark cores, to ensure low influence of background tasks. Cgroups also nicely interact with
the real time scheduler, as described here https://www.linuxjournal.com/article/10165, because
they allow to adapt the scheduler to ignore the other cores in its decision making process.
Note the exclusive cpu groups in this output:
```shell script
florian@bananapim3:~$ cset set
cset: 
         Name       CPUs-X    MEMs-X Tasks Subs Path
 ------------ ---------- - ------- - ----- ---- ----------
         root        0-7 y       0 y   116    2 /
         user        1-7 y       0 n     0    0 /user
       system          0 y       0 n    58    0 /system
```

Create a file called 'setup_cgroups.sh' and modify it with 'chmod +x setup_cgroups.sh':
```shell script
#!/bin/bash

sudo cset shield --cpu=1-7 -k on
```

This will isolate cores 1 to 7 for our benchmarks. To run the benchmarks on these cores use the following
or a similar command: `sudo chrt --fifo 90 cset shield -e --user=<user> <executable> \-- <arguments>`


### CPU frequency

Limiting the frequency to 1GHz makes sure that the banana PI dose not throttle during the tests.
Additionally, disabling any dynamic frequency scaling makes tests more reproducable.

Create a file called 'setup_cpu.sh' and modify it with 'chmod +x setup_cpu.sh':
```shell script
#!/bin/bash

echo "Writing frequency utils settings file..."
echo "ENABLE=true
MIN_SPEED=1412000
MAX_SPEED=1412000
GOVERNOR=performance" > /etc/default/cpufrequtils

echo "Restarting frequency utils service..."
systemctl restart cpufrequtils

echo "Done!"
echo "Try ./watch_cpu.sh to see  if everything worked."
echo "Test your cooling by stressing the cpu and watching the temperature output."
```

Create a file called 'watch_cpu.sh' and modify it with 'chmod +x watch_cpu.sh':
````shell script
#!/bin/bash

echo "Min/Max Frequencies"
cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_min_freq
echo "-----"
cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_max_freq

echo "Scaling Min/Max Frequencies"
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_min_freq
echo "-----"
cat /sys/devices/system/cpu/cpu*/cpufreq/scaling_max_freq

echo "Actual Frequencies"
cat /sys/devices/system/cpu/cpu*/cpufreq/cpuinfo_cur_freq

echo "Temps.."
cat /sys/class/thermal/thermal_zone*/temp
````

***BEFORE TESTS***:
To setup the CPU run ***`sudo ./setup_cpu.sh`*** before your tests. To see that the change worked
and the temperatures hold stable use the `./watch_cpu.sh` script.

### Map interrupts to core 0

Interrupts can infer with our benchmarks. We therefore map them to core 0 if possible and run our tests on
cores 1 to 7.

Create a file called 'map_interrupts_core_0.sh' and modify it with 'chmod +x map_interrupts_core_0.sh':
```shell script
#!/bin/bash

echo "Try to map interrupts to core 0."
echo "Some might fail because they can not be  mapped (e.g. core specific timers)."
echo ""
echo ""

echo 1 > /proc/irq/default_smp_affinity
for dir in /proc/irq/*/
do
	echo "Mapping $dir ..."
	echo 1 > $dir/smp_affinity
done
```

***BEFORE TESTS***: map the interrupts to core 0 using ***`sudo ./map_interrupts_core_0.sh`***

### Full time slices to RT scheduler

The RT scheduler in linux by default leaves some fraction of its scheduling time to non RT processes,
leaving the system in a responsive state if a RT application eats all CPU. We do not want this, as we
try to get a very predictable behavior in our RT scheduler.

Create a file called 'setup_rt.sh' and modify it with 'chmod +x setup_rt.sh':
```shell script
#!/bin/bash

sysctl -w kernel.sched_rt_runtime_us=1000000
sysctl -w kernel.sched_rt_period_us=1000000
````

***BEFORE TESTS***: give full time slices to RT tasks ***`sudo ./setup_rt.sh`***

## Running Tests

***Before running tests make sure to run the following scripts:***
- `sudo ./setup_cpu.sh`
- `sudo ./map_interrupts_core_0.sh`
- `sudo ./setup_rt.sh`

To run the tests use the following (or a similar command with different rt policy):

`sudo chrt --fifo 90 cset shield -e --user=<user> <executable> \-- <arguments>`

This maps the process to all cores but core 0 and runs them using the desired real time schedule and priority.

We found that interactive sessions can cause huge latency spices even with this separation,
therefore we advise on starting the benchmarks and then leaving the system alone until they are done.