UnixBench

UnixBench is a long-standing benchmark for Unix-like systems, originally from BYTE magazine and now community-maintained. It measures the overall performance of a Unix or Linux system rather than any single subsystem, which has made it a common quick check for comparing servers, virtual machines, and cloud instances despite its age.

The suite runs a collection of small tests covering integer and floating-point arithmetic (Dhrystone and Whetstone), process creation, pipe throughput, context switching, shell script execution, system-call overhead, and file copy at several buffer sizes. Each result is compared against a fixed baseline system, and the ratios are combined into an aggregate index.

What It Measures

The headline output is the System Benchmark Index Score, a single number summarizing relative performance against the reference machine. Underlying it are individual scores: Dhrystone (integer), Whetstone (floating point), process creation rate, pipe and context-switch throughput, shell parallelism, system-call overhead, and file-copy rates. UnixBench reports both single-CPU and multi-CPU (parallel-copy) indexes.

Methodology

Each test runs for a fixed duration, and its raw throughput is divided by the baseline system's result to produce an index value where the baseline scores a defined reference figure. The geometric mean of all index values forms the overall score. The multi-CPU run launches as many parallel copies as there are cores, so comparing single- and multi-CPU scores reveals parallel scaling. The suite compiles from source with the system compiler.

How to Interpret Results

Use the single-CPU index for per-core performance and the multi-CPU index for aggregate throughput; the ratio between them shows how well the system scales across cores. UnixBench is most useful for quick relative comparison of similar systems — for example ranking cloud instance types. Inspect individual subtest scores when one metric matters, such as process creation for fork-heavy workloads.

Limitations

UnixBench is old, and several tests (Dhrystone, Whetstone) are no longer representative of modern CPU behavior or compiler optimization, which can distort results. The aggregate index obscures which subsystem actually limits a workload. Results are sensitive to compiler version and flags, and the suite does not exercise networking, large-memory, or modern multi-core scaling realistically. Treat it as a rough comparative indicator, not an authoritative performance measure.