Skip to content

test/bench: the one real WAN benchmark is not reproducible from the repo — reconstruct + commit an automated harness #109

Description

@vxfemboy

crates/yip-bench/wan/results/readme-snippet.md credits wan/collect.sh as the generator, but that script does not exist in the tree, and the whole wan/ dir is untracked (with an SSH-key-bearing wan.env). Nobody can regenerate the one real cross-internet data point. This is the 'automated real testing/benchmark' ask. Plan: reconstruct + commit a scrubbed collect.sh (yip vs WireGuard vs OpenVPN: latency, TCP/UDP iperf3, and a netem loss-resilience sweep — the FEC differentiator); store SSH creds as GH Actions secrets (not a local wan.env); add a workflow_dispatch (or low-frequency scheduled) job that SSHes the two hosts and appends to a tracked wan/results/ history — turning today's one-off snapshot into a trend line. A fresh 2026-07-25 data point (current code, Utah<->LV, ~23ms): latency parity (23.7 vs 23.2ms), TCP 142 Mbit/s single-core single-stream, and 5% underlay loss -> 0.5% tunnel loss (FEC ~11x). Also: yip-bench/README.md still headlines a stale RaptorQ number (~18x off, codec removed in #50).

See docs/2026-07-25-project-review-and-roadmap.md for the full review synthesis.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions