Skip to main content

8: Optimizing canisters

Beginner
Motoko
Tutorial

Overview

Optimizing canisters is an important development step that can assure your canister is using the most efficient workflow by optimizing the canister's cycles usage and binary size. This guide covers how to optimize Motoko canisters.

Prerequisites

Before getting started, assure you have set up your developer environment according to the instructions in the developer environment guide.

Using wasm-opt

Wasm-opt is a general purpose Wasm optimizer that is now available in dfx, versions 0.14.0 and newer.

Wasm-opt can be used to enable canister optimizations through a configuration option in the project's dfx.json file, such as:

{
"canisters": {
"my_canister": {
"optimize": "cycles"
}
}
}

Optimization levels for cycle usage:

Using the "optimize": "cycles" option, you can expect a rough estimate of decreased cycles usage for Motoko canisters by around 10%.

The "optimize": "cycles" option is the recommended default, as it maps to optimization level 3 in the wasm-opt package.

The optimization levels for cycles usage are as follows:

O4
O3 (equivalent to “cycles”)
O2
O1
O0 (performs no optimizations)

Optimization levels for binary size:

To optimize the binary size instead, you can use the "optimize": "size" option. By using the size option, binary sizes can be reduced by roughly 16%.

The optimization levels for binary size are as follows:

Oz (equivalent to “size”)
Os

Each optimization preserves the Internet Computer specific metadata sections of each canister.

Note that in certain cases the optimizations can increase the complexity of certain functions in your Wasm module such that they are rejected by the replica. If you run into this issue, it is recommended to use a less aggressive optimization level such that you do not exceed the complexity limit.

More information on canister optimization and information on wasm-opt benchmark testing can be found on this forum post.

Next steps

Next, let's take a look at importing library modules.