throughput

TCP throughput sample

The guest connects to -addr, writes fixed-size chunks for -secs seconds, and prints the bytes sent and elapsed time. It needs a raw TCP sink that accepts bytes until the guest closes. It is not an iperf protocol client.

Build

From the repository root:

make wasm SAMPLE_DIR=local/wasm_samples/go/throughput

Local example

Use an already configured local dispatcher and executor. For the simplest setup, configure the executor to use the fallback packet counter. In a separate terminal, run a loopback sink with socat:

socat -u TCP-LISTEN:8080,bind=127.0.0.1,reuseaddr,fork OPEN:/dev/null

Then submit to an ID from dbl nodes:

dbl --timeout 30s run \
  --wasm local/wasm_samples/go/throughput/debuglet.wasm \
  --executor EXECUTOR_ID --allow 127.0.0.1 \
  --floor-bps 20000 --ceil-bps 100000 --duration 25s --wait \
  -- -addr 127.0.0.1:8080 -secs 20 -chunk 128

Replace EXECUTOR_ID; add --endpoint before the command if needed. Stop the sink when finished. Choose a guest duration shorter than the job budget and a client timeout long enough to observe completion.

A chunk larger than 8192 bytes is split across host calls, so -chunk changes how many calls the guest makes rather than how much reaches the sink.

The output is application-level throughput for this setup. It does not establish eBPF packet enforcement or link capacity. Kernel measurements require an isolated test network, suitable privileges, and separate packet-level verification.