Skip to content
Snippets Groups Projects
Select Git revision
  • master default protected
  • update-docker-compose-rpc-squid-names
  • network/gdev-800 protected
  • fix-252
  • 1000i100-test
  • hugo/tmp-0.9.1
  • hugo/distance-pool-index
  • network/gdev-803 protected
  • hugo/endpoint-gossip
  • network/gdev-802 protected
  • hugo/distance-precompute
  • network/gdev-900 protected
  • tuxmain/anonymous-tx
  • debug/podman
  • hugo/195-doc
  • hugo/195-graphql-schema
  • hugo-tmp-dockerfile-cache
  • release/client-800.2 protected
  • release/runtime-800 protected
  • feature/show_milestone
  • gdev-900-0.10.0 protected
  • gdev-900-0.9.2 protected
  • gdev-800-0.8.0 protected
  • gdev-900-0.9.1 protected
  • gdev-900-0.9.0 protected
  • gdev-803 protected
  • gdev-802 protected
  • runtime-801 protected
  • gdev-800 protected
  • runtime-800-bis protected
  • runtime-800 protected
  • runtime-800-backup protected
  • runtime-701 protected
  • runtime-700 protected
  • runtime-600 protected
  • runtime-500 protected
  • v0.4.1 protected
  • runtime-401 protected
  • v0.4.0 protected
  • runtime-400 protected
40 results
Blame Permalink
weights-benchmarking.md 2.92 KiB

Weights benchmarking

What is the reference machine?

For now (09/2022), it's a Raspberry Pi 4 Model B - 4GB with an SSD connected via USB3.

To cross-compile the benchmarks binary for armv7:

./scripts/cross-build-arm.sh --features runtime-benchmarks

The cross compiled binary is generated here: target/armv7-unknown-linux-gnueabihf/release/duniter

How to benchmarks weights of a Call/Hook/Pallet

  1. Create the benchmarking tests. See commit f5f2ae96 for a complete real example.

  2. Run the benchmark test on your local machine:

cargo test -p <pallet> --features runtime-benchmarks
  1. If the benchmark tests compiles and pass, compile the binary with benchmarks on your local machine:
cargo build --release --features runtime-benchmarks
  1. Run the benchmarks on your local machine (to test if it work with a real runtime). See 0d1232cd for a complete real example. The command is:
duniter benchmark pallet --chain=CHAINSPEC --steps=50 --repeat=20 --pallet=<pallet> --extrinsic=* --execution=wasm --wasm-execution=compiled --heap-pages=4096 --header=./file_header.txt --output=./runtime/common/src/weights/

  1. Use the generated file content to create the WeightInfo trait and the () dummy implementation in pallets/<pallet>/src/weights.rs. Then use the WeightInfo trait in the real code of the pallet. See 62dcc17f for a complete real example.

  2. Redo steps 3. and 4. on the reference machine.

  3. Use the runtime/common/src/weights/pallet_<pallet>.rs generated on the reference machine in the runtimes configuration. See af62a3b9 for a complete real example.

Note 1: Use relevant chainspec for the benchmarks in place of CHAINSPEC. For example --chain=gdev-benchmark has already created identities that can be confirmed by pallet identity

Note 2: If the reference machine does not support wasmtime, you should replace --wasm-execution=compiled by --wasm-execution=interpreted-i-know-what-i-do.

Generate base block benchmarking

  1. Build binary for reference machine and copy it on reference machine.
  2. Run base block benchmarks command:
duniter benchmark overhead --chain=gdev-benchmark --execution=wasm --wasm-execution=compiled --weight-path=./runtime/common/src/weights/ --warmup=10 --repeat=100
  1. Commit changes and open an MR.

Generate storage benchmarking

  1. Build binary for reference machine and copy it on reference machine.
  2. Copy a DB on reference machine (on ssd), example: scp -r -P 37015 tmp/t1 pi@192.168.1.188:/mnt/ssd1/duniter-v2s/
  3. Run storage benchmarks command, example:
duniter benchmark storage -d=/mnt/ssd1/duniter-v2s/t1 --chain=gdev --mul=2 --weight-path=. --state-version=1
  1. Copy the generated file paritydb_weights.rs in the codebase in folder runtime/common/src/weights/.
  2. Commit changes and open an MR.