BasisRust

Rust port of the Basis server and a headless Basis-compatible workload client. Much of the initial port was translated from C# with GPT assistance and then adapted and tested. Protocol parity and measured behavior remain the criteria for changes.

Prerequisites

• Rust/Cargo 1.95.0 or newer, installed with rustup. The server core declares Rust 1.95; the Docker builder pins 1.95.0.
• A native C/C++ compiler and linker for Cargo's native dependencies (on Debian/ Ubuntu, build-essential and pkg-config). GPU support is opt-in.
• Python 3 for the portable workload runner; Linux taskset only for CPU affinity.
• ffmpeg on PATH for the client's default voice re-encoding path. Avatar-only workloads do not require it. The C# interop harness separately requires .NET 8.

Build and run

From a fresh checkout, build the two workspaces independently:

CARGO_BUILD_JOBS=2 cargo build --locked --release --manifest-path BasisRustServer/Cargo.toml -p basis-server-console
CARGO_BUILD_JOBS=2 cargo build --locked --release --manifest-path BasisRustClient/Cargo.toml

Start the server from the repository root:

BasisRustServer/target/release/basis-server-console --base-dir BasisRustServer --config config/config.xml

In another terminal, run the client from its directory so its tracked sample Config.xml is found:

cd BasisRustClient
cargo run --locked --release

These commands use the compatibility password default for local testing. For deployment, configure a non-empty password on both ends; see configuration and Docker deployment.

Layout and reference

• Architecture: server workspace/domain crates and shared client dependencies.
• Server and client: CLI, configuration and optional features.
• Protocol and ACK vectors, moderation parity, and verified audit.
• Fresh workload runs and published results, historical artifact availability.
• Contributing, security guidance, MIT license.

Both BasisRustClient/Config.xml and BasisRustServer/Config.xml are tracked samples. Ignore rules protect untracked local Config.xml, config/, captures, logs and build directories; they do not hide edits to tracked samples. Keep local credentials in ignored files or runtime environment settings. Each workspace has its own Cargo.toml, lockfile and target/ output; no root Cargo workspace exists.

Performance evidence

Rayon currently executes parallel receiver builds. In a four-run ABBA 2,000-client comparison, reducing the minimum batch size from 16 to 4 lowered mean observer p95 update gaps from 1,102 to 851 ms while server CPU rose from 325% to 458%. Those two repeats per variant on one host show a cadence/CPU tradeoff in a synthetic loopback workload. They do not establish that Rayon causes most overhead or that replacing it with a custom executor would improve performance.