[limonuser]

This is the first part of #98: a cell that replays heavy PDOs (mahimahi style traces), following the design @Lethe10137 described there. It uses the pdo-model feature and PDO trace types from stack-rs/netem-trace#29, released in netem-trace 0.5.0.

What it does:

• PdoReplayCell takes a HeavyPdoTrace ((interval since last PDO, bytes)) and a packet queue. No bandwidth anywhere
• The bytes of a PDO go to the packet at the head of the queue, counted by L3 length like mahimahi. A big packet can take several PDOs (it keeps the bytes it already got), and one PDO can send several small packets
• dequeue() returns one packet at a time. The bytes left in the current PDO stay in the cell, so the next dequeue() sends the next packet right away without sleeping
• If nothing is queued, the bytes of a PDO are wasted. While the cell is idle it doesn't wake up for every PDO, it just skips the ones from before the next packet arrived
• When the trace ends nothing more is sent. For mahimahi files load_mahimahi_heavy_pdo_trace repeats forever by default anyway
• With a zero buffer queue, a packet that arrives while another one is waiting for a PDO is dropped
• Small helpers on AQM: is_empty() and is_queue_empty()

Not in this PR (next ones):

• config wiring, so it can be used from a config file with a mahimahi trace
• the CellReplay light PDO part (bypass packets, base delay, the inactive timer)

Things I wasn't sure about:

• I count bytes with l3_length() to match mahimahi, not with the queue's get_packet_size() (which can add 14 B for L2). Should it follow the queue's BwType instead?
• Leftover bytes of a PDO are dropped as soon as the queue is empty, they're never saved for later. I think that's what you meant in Add functionality to replay cellular traces #98 but tell me if not

How I tested (Ubuntu 24.04 in WSL, stable Rust, with netem-trace 0.5.0):

• 8 new tests in cells::pdo: bytes carrying over, many small packets in one PDO, wasted bytes after being idle, a mahimahi trace with two lines in the same ms, zero buffer, and a random test (3 seeds) that checks every departure time against a small copy of mahimahi's use_a_delivery_opportunity logic
• I broke the "keep leftover bytes" line on purpose to check that the tests catch it (6 of 8 failed). That's also how I found the test could hang forever, so replay() now fails after 5 s
• cargo clippy --all-features --all-targets -- -Dwarnings and cargo fmt --all -- --check are clean
• 3 old timing tests fail on my machine (zero_buffer_bw_replay, test_replay_delay_cell, test_replay_delay_cell_change_state), but they fail the same way on main without this change, so I think it's my WSL setup

Refs #98

🤖 Generated with Claude Code

[limonuser]

netem-trace#29 is merged and 0.5.0 is out, thanks @Lethe10137 and @BobAnkh for the reviews!

I pushed 88eb4e7, which bumps netem-trace to 0.5.0 (no local patch anymore), so CI should be able to build this now. On my machine the 8 cells::pdo tests, clippy and fmt all pass with 0.5.0.

I'm moving this out of draft. The two questions from the description are still open: should bytes be counted with l3_length() or follow the queue's BwType, and is it right to drop leftover PDO bytes when the queue is empty? Once those are settled I'll start on part 2, the config wiring.

[Lethe10137]

netem-trace#29 is merged and 0.5.0 is out, thanks @Lethe10137 and @BobAnkh for the reviews!

I pushed 88eb4e7, which bumps netem-trace to 0.5.0 (no local patch anymore), so CI should be able to build this now. On my machine the 8 cells::pdo tests, clippy and fmt all pass with 0.5.0.

I'm moving this out of draft. The two questions from the description are still open: should bytes be counted with l3_length() or follow the queue's BwType, and is it right to drop leftover PDO bytes when the queue is empty? Once those are settled I'll start on part 2, the config wiring.

• For the BwType: just follow the queue's BwType. And it should be defaulted to be l3_length() as a convention in Rattan
• Yes, clear the PDO bytes when the queue is empty.

[limonuser]

Thanks @Lethe10137! I pushed 4a697d2. Now the cell counts each packet as l3_length() plus the extra length from the queue's BwType, so by default it's still just the L3 length (NetworkLayer), and I added a test with LinkLayer. The leftover PDO bytes were already dropped when the queue is empty, so nothing changed there.

I'll start on part 2 (the config wiring) next.