Skip to content

Kernel compare - #8

Merged
mp0rta merged 11 commits into
mainfrom
kernel-compare
Oct 12, 2026
Merged

mp0rta merged 11 commits into
mainfrom
kernel-compare

Conversation

@mp0rta

@mp0rta mp0rta commented Oct 12, 2026

Copy link
Copy Markdown
Owner

No description provided.

mp0rta added 11 commits October 12, 2026 05:30
INTEROP_CLIENT=kernel runs kclient.py (Linux MPTCP, 10.0.1.1/10.0.2.1 with
a leg-B endpoint) instead of mptcp-rs-tun; INTEROP_BENCH=1 gives both
clients the same conditions: offloads off and gso_max_segs 1 on the shaped
veths, no TUN capture, check.py logs-only plus an MP_JOIN check, and a
steady-window CPU sample (cpu.py). m2-unshaped: A+B without netem, capture
or byte verification.

net:net_dev_queue max skb len per dev, bench m2-goodput, 20 s mid-transfer:
- kernel c2s, gso_max_segs 1: a0 1514, b0 1514 (a1/b1 ACKs 78)
- kernel s2c, gso_max_segs 1: a1 1514, b1 1514 (a0/b0 ACKs 78)
- kernel c2s, device GSO off but no gso_max_segs: a0 44234, b0 11482
- mptcp-rs c2s, no gso_max_segs: a0 1514, b0 1514
A saturated kernel sender spends ~97 % of its core in sys time (the kernel
stack's send path, softirqs included), which a native sender pays too.
@mp0rta
mp0rta merged commit 8930d97 into main Oct 12, 2026
14 checks passed
@mp0rta
mp0rta deleted the kernel-compare branch October 12, 2026 10:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant