Summary v0.9.27
Based on our detailed analysis, we have compiled polkadot version 0.9.27 by modifying the production profile in the Cargo.toml root folder of the polkadot source code.
In the table below, the benchmark scores (higher is better) and the timing of the Remark Extrinsic (lower is better) are listed for a few machines with different CPU architectures. Please look here on why we only focus on these values and how many repetitions were done to reduce the statistical errors.
The tables below are not well visible on mobile. Please use the desktop version.
The column impr. (%) is the relative improvement in percentage compared to the official polkadot binary on the same machine.
While not very powerful nor recent, this cpu is quite popular. It benefits a lot from optimization: 25% and 10% improvement for SR25519-Verify and Extrinsic timing, resp., at the expense of -1% for BLAKE2-256.
| build | BLAKE2-256 | impr. (%) | SR25519-Verify | impr. (%) | Extr-Remark | impr. (%) | toolchain | codegen-units | lto | opt-level |
|---|---|---|---|---|---|---|---|---|---|---|
| 4 | 994.2 | -1.6 | 697.4 | 28.2 | 85519 | 12.6 | stable | 1 | fat | 2 |
| 7 | 1003.5 | -0.6 | 684 | 25.7 | 86683 | 11.4 | stable | 1 | thin | 3 |
| official | 1010 | 544.1 | 97889 | nightly | 16 | thin local | 3 | |||
| docker | 972.1 | -3.7 | 531 | -2.4 | 100515 | -2.7 | nightly | 16 | thin local | 3 |
| production | 959.7 | -5 | 703.1 | 29.2 | 85197 | 13 | nightly | 1 | fat | 3 |
A very recent cpu with excellent performance for the official binary. Optimization improves all the numbers but the effect is limited: about 2% to 5%.
| build | BLAKE2-256 | impr. (%) | SR25519-Verify | impr. (%) | Extr-Remark | impr. (%) | toolchain | codegen-units | lto | opt-level |
|---|---|---|---|---|---|---|---|---|---|---|
| 38 | 1400 | 4.5 | 1020 | 1.9 | 54505 | 1.4 | nightly | 1 | fat | 2 |
| 45 | 1400 | 4.5 | 1022.9 | 2.1 | 54256 | 1.8 | nightly | 16 | fat | 3 |
| 14 | 1400 | 4.5 | 1020 | 1.9 | 55013 | 0.5 | stable | 1 | fat | 2 |
| official | 1340 | 1001.4 | 55263 | nightly | 16 | thin local | 3 | |||
| docker | 1340 | 0 | 1002.6 | 0.1 | 55423 | -0.3 | nightly | 16 | thin local | 3 |
| production | 1350 | 0.7 | 999.6 | -0.2 | 55317 | -0.1 | nightly | 1 | fat | 3 |
An old Xeon from 2014 that would not be adequate for a polkadot validator. Optimization leads to improved performance in all three scores of about 5% to 15%.
| build | BLAKE2-256 | impr. (%) | SR25519-Verify | impr. (%) | Extr-Remark | impr. (%) | toolchain | codegen-units | lto | opt-level |
|---|---|---|---|---|---|---|---|---|---|---|
| 3 | 711.4 | 5.3 | 479.5 | 17 | 129757 | 3.9 | stable | 1 | thin | 3 |
| 9 | 687.4 | 1.7 | 479.2 | 16.9 | 124197 | 8 | stable | 16 | fat | 3 |
| official | 675.7 | 409.9 | 135009 | nightly | 16 | thin local | 3 | |||
| production | 692.1 | 2.4 | 473.4 | 15.5 | 124522 | 7.8 | nightly | 1 | fat | 3 |
In order to generate an optimal build, you need to modify the Cargo.toml file as follows:
[profile.production]
inherits = "release"
codegen-units = CHANGE THIS
lto = CHANGE THIS
opt-level = CHANGE THIS
where CHANGE THIS determines the build profile specified above. Afterwards, you build as usual:
rustup override set nightly
export RUSTFLAGS="-C target-cpu=native"
cargo build --profile=production --target=x86_64-unknown-linux-gnu --locked -Z unstable-options
Instead of manually building, you can also use our convenient Python script. It allows you to build several binaries in one go and specify options in a simpler way.
If you want to contribute to this list:
- Take a look here to install the necessary packages.
- Then you run our benchmark script when there are no other resource hungry tasks running.
- Finally, send us the
featherfiles generated by the parse script.