#####################################################################
May 15, 2025
Attendees
1. Patrick Robb
2. Paul Szczepanek
3. Luca Vizzarro
4. Manit Mahajan
5. Aaron Conole
#####################################################################
Minutes
=====================================================================
General Announcements
* DPDK Summit Prague was held May 8-9
* From talking with Shai (AWS lab maintainer) at DPDK Summit, it sounds like he has an interest in the CI meetings, but that they are not a good time for him. Patrick should find one of those “free timeslot polling” tools and send such a poll to all the maintainers to figure out a timeslot which would work for everyone.
=====================================================================
CI Status
---------------------------------------------------------------------
UNH-IOL Community Lab
* Pcapng unit test timestamp error:
* This is a known issue which is coming from how the unit test leverages the TSC timestamp register. We had conducted a simple test on this last year which confirmed that the TSC did not run backwards, but the conversation petered out. There is another failure now, so we need to take another look at this.
* Does TSC wraparound?
* Email: https://inbox.dpdk.org/ci/CAJvnSUCXns=JXV0FsWENqsifNzCA2NQhqJBCwDsfh=S4BbWy2A@mail.gmail.com/T/#t
* Bruce meson version patchseries:
* Adds a script which reports the minimum meson version for a given DPDK version, and also adds that into the .ci/linux-setup.sh script
* Bruce is suggesting some changes for where the LTS branches live:
* Currently on github:
* DPDK Repo: contains DPDK and the tagged releases (get no backports)
* DPDK-stable repo: Contains the LTS releases which get backports
* Bruce proposes to bring the LTS releases with backports to the DPDK repo
* There should be no adverse affects on existing CI processes
* This should also encourage people to use the github DPDK repo more instead of git.dpdk.org. As a reminder, all ci labs should be doing this.
* Patrick is back from DPDK Summit starting today, and can start to address a few CI lab issues which have been backlogged since last week:
* MLX5 DTS func tests - After some reconfiguration right before DPDK summit, we started to rx no packets for a few testsuites. Patrick will investigate today and figure out what’s going on.
* E810 performance test issues:
* Cody was investigating this issue and he has a solution, but Patrick needs to approve the new changes and merge them today.
* New topology: 2 E810 NICs connected against 1 E810 DUT NIC.
* TREX change on the single core forwarding test: Trex (not testpmd or DUT E810) had an issue with sending rate given a certain mix of packets which are used by default in the single core forwarding testsuite. Cody has updated the testsuite to use a uniform packet model (instead of a mix), which resolves the issue on the traffic generator side. The performance on the DUT is still good (like it always was) but it is no longer affected by the traffic generator issue.
* Patrick will review this solution today - most likely we can re-enable the test
---------------------------------------------------------------------
Intel Lab
* None
---------------------------------------------------------------------
Github Actions
* None
---------------------------------------------------------------------
AWS
* We met with Shai at DPDK Prague. Some takeaways:
* He is interested in supporting the recheck framework, and would like to sync with us on that at an upcoming meeting.
* But we will have to wait 2 weeks for that, so Patrick might as well start the conversation over email
* DTS: For sending packets between TG and DUT AWS EC2 instances, we cannot send with a mac destination address - we have to send with an IP destination address.
---------------------------------------------------------------------
Loongarch Lab
* None
=====================================================================
DTS Improvements & Test Development
* DPDK-Pktgen:
* The Suricata guys got to chatting with Patrick at DPDK Summit about their DPDK-pktgen usage. They are able to use it for 100gbps+ TG operations, and they have shared some code showing how they use dpdk-pktgen in their CI testing. Patrick can share with the UNH team, and can probable also overview what they shared at the DTS meeting next week.
* Flow checker tool:
* There was a flow rule checker tool for DPDK 22.11 which is no longer being maintained publicly, but the original creator still maintains it privately for his usage, and can push the updates if we need.
* Patrick will look at this tool from 22.11
* Rte_flow: MVP testsuite is done - he is submitting to patchwork this morning.
* Birds of a Feather:
* This was productive and there were some good suggestions:
* Suggestions:
* Code coverage reports for DTS
* Support smaller topologies: namely testing DTS with 1 dual port NIC connected in loopback.
* Or, run DTS from a laptop using virtual ports
* Write some blogs about how to use DTS this year
* Support other programs (user apps)
* Reviews:
* Patrick will send his review of the TREX patchseries today
* Luca reviewed the l4 port matching series. They are conceptually fine, but he had a few comment or docstring nits which he left. Dean can update this today and resubmit.
* Improve port handling series:
* Groups all the port handling under the topology class - the topology class becomes the interface for port handling.
* Patrick and UNH guys need to review this
* VF patchseries:
* It would be better if these were integrated in the topology.
* May have to make the next version a depends-on the “improve port handling” series
* Confirmed that bnxt_en has the same promiscuous mdoe support requirement as Intel (bind to kernel driver, then set “trust” mode on the port, then bind to vfio-pci).
* Patrick merged the port_stats testsuite, but it includes an error for building the docs. There is a simple fix which Luca suggested, and Patrick will submit it.
=====================================================================
Any other business
* Next Meeting May 29, 2025