Subtrace is incredibly lightweight and adds negligible overhead to the latency of your backend requests. Even under extreme load, it adds less than 1 millisecond in latency; under typical loads, you can expect less than 0.1ms in added latency. For reference, the latency between machines within the same datacenter is usually 4-10 milliseconds. We periodically and publish measure the following latency and throughput impact of Subtrace in the next two graphs. These benchmarks were performed on a AWS m5a.4xlarge machine with 16 vCPUs and 64 GiB memory. Both the client and server were on the same machine in order to eliminate network noise. In each benchmark, 16 parallel clients continuously sent requests to a server running running under Subtrace for 10 seconds. While these benchmarks don’t represent all types of workloads, they should give you a sense of what to expect from integrating Subtrace. We encourage everyone to measure the impact of Subtrace on their own app. If these numbers aren’t aligned with your observations, let us know and we’ll investigate. We care about performance deeply.