stormshield-fabs

Taking inspiration from https://github.com/mozilla/neqo/issues/1472, we've been plotting sent, acked and lost packets over time (on a simulated network with a 10 ms latency, `--upload-size 1000000000 --duration 20`). So far, all...

Comparing the timings of packet emission, reception and acknowledgment on both the server and the client showed that the packet losses were real. We are using a simulated network using...

More tests on the internet link: - baseline (32MB for `stream_receive_window` and `send_window`, recv buffer size at 425984) : 363 MBps - with recv buffer size at 2097152 : 1.03...

> You previously excluded the socket buffer difference hypothesis Those numbers were correct : with the default `rmem_max=212992`, both `quinn/perf` and `msquic/secnetperf` use a receive buffer size of `425984` for...

We were able to plot the socket's receive buffer size over time by parsing `ss` output. (I haven't figured out a better way the retrieve this data yet). The figure...

Yes, the maximum execution times we collected for `EndpointDriver::poll` and `recvmmsg` were always very close.

We replicated the `recvmmsg` instrumentation in `msquic` and saw similar maximum latency of around 1ms. `msquic` appears to be using a dedicated thread for all socket reads and writes (at...

I encountered the same issue while using a [`[patch]` section](https://doc.rust-lang.org/cargo/reference/overriding-dependencies.html#the-patch-section). The following `Cargo.toml` ```toml [package] name = "eyre-bazel-demo" version = "0.1.0" edition = "2024" [dependencies] eyre = "1" [patch.crates-io] eyre...