Benchmarks
All benchmarks measured on the Rustbox platform infrastructure. Numbers represent end-to-end latency including sandbox creation, execution, and teardown.
Per-language latency
Section titled “Per-language latency”Median wall time for a minimal program (print("hello") equivalent) per language.
| Language | Median latency | Notes |
|---|---|---|
| Python | 36 ms | CPython interpreter startup |
| JavaScript | 36 ms | Bun runtime |
| TypeScript | 36 ms | Bun runtime (same as JS) |
| C | 60 ms | GCC compilation + execution |
| C++ | 72 ms | GCC compilation + execution |
| Go | 120 ms | Go compiler + execution |
| Rust | 140 ms | rustc compilation + execution |
| Java | 445 ms | javac + JVM startup |
Interpreted languages (Python, JS, TS) are fastest because there is no compilation step. Compiled languages include compilation time within the sandbox.
Throughput
Section titled “Throughput”Sustained throughput under load with concurrent submissions.
| Language | Executions/sec |
|---|---|
| Python | 148 |
| JavaScript | 104 |
| TypeScript | 104 |
| C | 85 |
| C++ | 72 |
| Go | 55 |
| Rust | 48 |
| Java | 18 |
Throughput is measured with sufficient concurrent clients to saturate the platform’s worker pool. Each execution is a full sandbox lifecycle (create, execute, teardown) with isolation verification.
What the numbers include
Section titled “What the numbers include”Every measurement includes the full cost:
- Sandbox creation (namespaces, cgroups, mount setup, seccomp filter installation)
- Credential drop and privilege stripping
- Compilation (for compiled languages)
- Execution
- Sandbox teardown and resource cleanup
There is no warm-start optimization, no container reuse, no pre-created pools. Every execution builds a fresh sandbox from scratch.
Machine specifications
Section titled “Machine specifications”Benchmarks were measured on the following hardware:
| Spec | Value |
|---|---|
| CPU | AMD EPYC (4 cores) |
| Memory | 16 GB |
| Kernel | Linux 6.x (cgroup v2) |
| Storage | NVMe SSD |
Production infrastructure may differ. Reach out if you need latency guarantees for your use case.