I plan to discuss those on VPP semimonthly meeting in few minutes, this is both my agenda and a summary for people not joining that meeting.
RC1 is this week and some issues are quite bad and may warrant some reverts, but as bugfixes are allowed until RC2 we have still few weeks time. 1. QAT no traffic after DPDK bump [0]. My bad, I got distracted by transient issues and forgot to re-run larger test selection. I am still actively investigating, I have put my findings so far into a quick ticket [1]. Kai was on CSIT meeting and he mentioned some compile flag for symmetric ciphers, but currently plugins/dpdk/cryptodev/ registers zero handlers, which cannot be correct. 2. Before that, OpenSSL bump [2] had introduced an IPsec HW regression on 3n-srf. Both issues are visible in trending [3]. 3. New packed/natural buffer layout [4], on Xeon with huge pages it causes a regression. I opened a ticket [5] in favor of keeping old behavior (all natural) as default [6], but if most of users like the new default I can adapt CSIT to set “natural” in startup.conf explicitly instead. 4. Octeon driver got some merges, but now VPP usually [7] does not start [8] in CSIT. I am not sure if it is an infra issue or VPP source issue. I will try to investigate before RC2. Vratko. [0] 46433: dpdk: bump to DPDK 26.07 and rdma-core 64.0 | https://gerrit.fd.io/r/c/vpp/+/46433 [1] No cryprodev handler registers since DPDK bump to 26.07 | https://github.com/FDio/vpp/issues/3740 [2] 46462: build: update to openssl 4.0.1 in optional deps | https://gerrit.fd.io/r/c/vpp/+/46462 [3] https://csit.fd.io/trending/#eNrtVctOwzAQ_JpwQRvZSUzKoYeW_Ady7W1jESfGa6jC15PHwcmhEiDxUMXFjx2vZ7yjlSl0Hh8Jm20i9km5T7LS6GFI8t3tML06B3kLaD1wxk6YOY4bztQzaKefQPnehQ644JsDcAUYauMK4whVEdqmPkvqWwVD7CAJwbQBJFIm7k7KgvV-JMoeRiL9ElasEXF1H5HLWmKC9ChjxiwxogFpwfQZwfGOo5cWybxhvGgqQTyhhqouQLXmD71boKs61F2UJ0Y3OEtzlgqeFiyeisFsDIoqKasJ-lEDnf43cHqg_n4DyR__TAde0HL9HXj_i159sdmuzquPNtvCK1HdtJ230w83rKnuzhC8kQ1t56x5Mye8A2ma-UA [4] 46182: buffers: fix pool allocation on small pages | https://gerrit.fd.io/r/c/vpp/+/46182 [5] Packed buffer layout as new default has worse performance with huge pages in CSIT | https://github.com/FDio/vpp/issues/3739 [6] 46703: buffers: make natural layout always default | https://gerrit.fd.io/r/c/vpp/+/46703 [7] https://csit.fd.io/stats/#eNrLyk-yTS7OLNEtKyjQLUgtStPNLSrSTUnMzKnUzU0sLkkt0jXO081PLgEARDUPQQ [8] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-mrr-daily-master-3n-oct/35663534320/log.html.gz#s1-s1-s1-k1-k6
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#27198): https://lists.fd.io/g/vpp-dev/message/27198 Mute This Topic: https://lists.fd.io/mt/121375868/21656 Group Owner: [email protected] Unsubscribe: https://lists.fd.io/g/vpp-dev/leave/14379924/21656/631435203/xyzzy [[email protected]] -=-=-=-=-=-=-=-=-=-=-=-
