graphql-engine/server/benchmarks/README_AMI.md

We've set up an EC2 AMI to be our test runner, to be run on c4.8xlarge (the benchmark scripts depend on various details of this instance type; e.g. the number and arrangement of cores).

For now this is provisioned and modified manually. Here's a record of what we did:

sudo apt update
sudo apt dist-upgrade
sudo apt install cpupower linux-tools-common linux-tools-aws postgresql-client-common git
sudo apt install docker-ce docker-ce-cli containerd.io
sudo groupadd docker && sudo usermod -aG docker ${USER}
sudo apt install fio linux-tools-aws linux-tools-common
sudo apt install jq

Added following to /etc/rc.local:

#!/bin/bash

# Disable deeper sleep states, use "performance" governor:
cpupower frequency-set -g performance
cpupower idle-set -D10
# Other voodoo...
# https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/processor_state_control.html
# Disable turbo-boost for less variability
echo 1 > /sys/devices/system/cpu/intel_pstate/no_turbo

Edit /etc/default/grub and add intel_idle.max_cstate=1 idle=poll to GRUB_CMDLINE_LINUX_DEFAULT:

# https://engineering.mongodb.com/post/reducing-variability-in-performance-tests-on-ec2-setup-and-key-results
GRUB_CMDLINE_LINUX_DEFAULT="quiet splash intel_idle.max_cstate=1 idle=poll"

Then run:

$ sudo grub-mkconfig -o /boot/grub/grub.cfg 

Once we save the AMI, we give it the Name tag of hasura-benchmarks-runner.