[Lethe10137]

Groups the changes built around the trace start instant (TRACE_START_INSTANT):

• first-payload (feature first-payload): start traces on the first packet that carries payload, instead of the first packet. Includes fixes so the Bw cell and the token bucket cell honor first-packet / first-payload correctly.
• mimic mahimahi: a Delay cell mode that mimics mahimahi's behavior, plus assets/mimic_mahimahi.toml and a duration serde helper. It is implemented by align the packet's timestamp to the next ms at the ingress of delay cell. So that, if such delay cell is the last cell before the veth, packets will be sent out in per-ms bursts.
• drift log (features drift-log, drift-stat): log the trace start time and per-packet time drift in rattan-log, and output drift statistics as .jsonl (adds flowstats). Includes analysis scripts under scripts/ and docs in guide/src/ref/packet-log.md.

Stack (merge in order, rebase-merge; review the diff against the base branch). This is 2/5:

1. feat(loss): byte trigger for loss cells #217 feat(loss): byte trigger for loss cells
2. 👉 feat: first-payload start, ms granularity timestamp alignment, and drift log #218 feat: first-payload start, mahimahi-like delay, and drift log
3. perf(veth): update tokio and reduce ingress overhead #219 perf(veth): update tokio and reduce veth ingress overhead
4. fix(env): set up netns persist dir once per process #220 fix(env): set up the netns persist directory once per process
5. refactor(env): make Env generic and move I/O drivers into rattan-env #221 refactor(env): make Env generic and move I/O drivers into rattan-env

🤖 Generated with Claude Code

[BobAnkh]

Please split this pr into several ones: one PR to netem-trace, one only containing first-payload codes and one only containing drift log. And what is mahimahi-like delay?

[BobAnkh]

We should move this to netem-trace instead of here. We should form a new trait or a new wrapper struct

[Lethe10137]

Please split this pr into several ones: one PR to netem-trace, one only containing first-payload codes and one only containing drift log. And what is mahimahi-like delay?

Fine. For the "mahimahi-like delay" I meant aligning the packet's timestamp to the next ms at the ingress of a delay cell. So that, if such delay cell is the last cell before the veth, packets will be sent out in per-ms bursts. I changed that to "ms granularity timestamp alignment" -- I'm not sure if that is the correct name for the experimental feature.

[BobAnkh]

Please split this pr into several ones: one PR to netem-trace, one only containing first-payload codes and one only containing drift log. And what is mahimahi-like delay?

Fine. For the "mahimahi-like delay" I meant aligning the packet's timestamp to the next ms at the ingress of a delay cell. So that, if such delay cell is the last cell before the veth, packets will be sent out in per-ms bursts. I changed that to "ms granularity timestamp alignment" -- I'm not sure if that is the correct name for the experimental feature.

@Lethe10137 Yes, that's good, but then please also split it into another PR please. They are three different things.

[Lethe10137]

Please split this pr into several ones: one PR to netem-trace, one only containing first-payload codes and one only containing drift log. And what is mahimahi-like delay?

Fine. For the "mahimahi-like delay" I meant aligning the packet's timestamp to the next ms at the ingress of a delay cell. So that, if such delay cell is the last cell before the veth, packets will be sent out in per-ms bursts. I changed that to "ms granularity timestamp alignment" -- I'm not sure if that is the correct name for the experimental feature.

@Lethe10137 Yes, that's good, but then please also split it into another PR please. They are three different things.

I would split this PR in to several stacked PR, i.e. 2a-first-payload, 2b-drift-log, 2c-mimic-mahimahi, after https://github.com/stack-rs/netem-trace/pull/30/changes is merged.

[BobAnkh]

Please split this pr into several ones: one PR to netem-trace, one only containing first-payload codes and one only containing drift log. And what is mahimahi-like delay?

Fine. For the "mahimahi-like delay" I meant aligning the packet's timestamp to the next ms at the ingress of a delay cell. So that, if such delay cell is the last cell before the veth, packets will be sent out in per-ms bursts. I changed that to "ms granularity timestamp alignment" -- I'm not sure if that is the correct name for the experimental feature.

@Lethe10137 Yes, that's good, but then please also split it into another PR please. They are three different things.

I would split this PR in to several stacked PR, i.e. 2a-first-payload, 2b-drift-log, 2c-mimic-mahimahi, after stack-rs/netem-trace#30 (changes) is merged.

The merge of #30 is done, you can do the PR split now