hp1/hp3: persistent jumbo MTU 9700 on the 25 GbE RoCE links - #27
Open
randomizedcoder wants to merge 1 commit into
Open
hp1/hp3: persistent jumbo MTU 9700 on the 25 GbE RoCE links#27randomizedcoder wants to merge 1 commit into
randomizedcoder wants to merge 1 commit into
Conversation
Set enp1s0f0np0 / enp1s0f1np1 to MTU 9700 (ConnectX-4 Lx driver maxmtu is 9978; 9700 is NVIDIA's practical max for this NIC) via a systemd-udevd .link file, so it applies at device appearance independent of the imperative xdp2.testbed NIC wiring (whose `jumbo` option only hardcodes 9000). This lifts the RoCE path MTU (ibv active_mtu) from 1024 to its 4096 ceiling and raises the single-stream TCP baseline. Verified end-to-end: ping -M do -s 9672 passes 0% loss hp1<->hp3. Supports uds-rdma-proxy design 37 (large-frame throughput: a single stream now beats single-stream TCP). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Set the 25 GbE RoCEv2 links (
enp1s0f0np0,enp1s0f1np1) on hp1 and hp3 to a permanent MTU 9700 — the ConnectX-4 Lx driver'smaxmtuis 9978, and 9700 is NVIDIA's practical max for this NIC (MT27710).Done via a
systemd.network.links.linkfile (roce-mtu.nix) so it applies at device appearance through systemd-udevd, independent of the imperativexdp2.testbedNIC wiring (whosejumbooption only hardcodes 9000).Why
Lifts the RoCE path MTU (
ibv active_mtu) from 1024 → 4096 (its ceiling) — 4× fewer RoCE packets per frame — and raises the single-stream TCP baseline. This is the host-side half of uds-rdma-proxy design 37: with jumbo + large frames a singleurpstream now beats a single TCP stream (copy path 103% of jumboiperf2 -P1, zero-copy path 97.6% of 25 GbE line).Verified
ip -d link show enp1s0f0np0→mtu 9700;ibv_devinfo→active_mtu: 4096.ping -M do -s 9672 <peer>→ 0% loss end-to-end (jumbo path passes; if a future path can't, dropMTUBytesto"9000").🤖 Generated with Claude Code