Benchmarks

Benchmark

Orter is designed with performance as a primary focus. We continually optimize our system to deliver the fastest possible experience across a wide range of use cases. This page demonstrates how quickly Orter performs under different configurations and APIs, helping you understand the impact of various options and how Orter compares in real-world scenarios.

Latency

Comparison to state of the art

Methodology

The benchmark and worker are run on the same system. For Orter interfaces that support networking, localhost is used. The measured time is $T = T_{recv} - T_{send}$, where:

• $T_{send}$ is the moment the not-yet-serialized lookup request is passed to the client library for serialization and submission.
• $T_{recv}$ is the moment the client receives the already deserialized response from the server.

Therefore, $T$ includes some overhead from the serialization and deserialization processes.

Each run uses a $1\text{s}$ warmup period, followed by measurement of the average time over $10\text{s}$ of continuous requests.

On average $10\%$ of all requests generate a result (i.e., a successful lookup). The rest are lookups where no data is returned and therefore no response is returned by the worker.