[CSIT-1980] 2n-zn2: Regression in l2xcbase around 2024-10-10

[vvalderrv]

Description

It is suspicious other testbeds and tescases do not seem to be affected. Comparing show runtime before [0] and after [1] may explain ~6% performance decrease seen in comparison tables [2], and bisect runs up to [3] and perf verify [4] all identify 41168 [5] as the cause on master, but looking at trending [6] I am not convinced the change is real. I suspect this may be just some unexpected compiler sensitivity as in CSIT-1816.

[0] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2406-2n-zn2/59/log.html.gz#s1-s1-s1-s5-s17-t1-k2-k9-k9-k10-k1-k1-k1-k12

[1] https://logs.fd.io/vex-yul-rot-jenkins-1/csit-vpp-perf-report-iterative-2410-2n-zn2/30/log.html.gz#s1-s1-s1-s5-s17-t1-k2-k9-k9-k10-k1-k1-k1-k12

[2] https://csit.fd.io/comparisons/#eNqNkM0OgjAMx58GLgQjk48TB9GrF-ILLFAMEcbSDYI8vR0fTjyZLN3a_rq2fwUNFBrK1EkyhyUIFSCIAujtnM7eElWgV_-OPaUu3hw0lXUndmjZbyhdg5TGWXjKDIA2iY1i4TH2WXggi_QbV2DxWlTILc2EPwnms-gBTAbjOCTB0S9l-bQVRYefsSMzS3AxbnRd8xW2qp72SBxmO0brl9wTtzxfiWQ-W69WcvxDJaJ4C_p7cSvEggy86eFHF1qOdCH7pcvc3xUdtqkZLbq6Xa-bGlCt_hu0d3oy

[3] https://jenkins.fd.io/view/vpp/job/vpp-csit-bisect-master-ubuntu2204-x86_64-2n-zn2/15/

[4] https://jenkins.fd.io/job/vpp-csit-verify-perf-master-ubuntu2204-x86_64-2n-zn2/13/

[5] 41168: dpdk: xstats as symlinks | https://gerrit.fd.io/r/c/vpp/+/41168

[6] https://csit.fd.io/trending/#eNp1kMEOgjAMhp9mXkwTN52cPIi8hxlbFeKAZZsIPL2DmBQOXra139-_XUPsPN4D2guTOctyJrLapIMdr_t09c6BaGFqBQj5ROH4MPQZP4Bx5gVWwPlUAteAsUrRoEsVEBrvZwdxmx3MO27siLhqJPK3CemVR0UFVhCJGFZN1rOQ5uFVg6GekIRpduI67YEQ11v3OLoV_X0wK5gsdm3nm2V56R2q7gPR18qGJZWKl2BWy-IL57FjyA

Assignee

Vratko Polak

Reporter

Vratko Polak

Comments

• vrpolak (Tue, 19 Nov 2024 13:39:20 +0000): Vhost test also show similar regression in trending [7], twice, also confirming compiler quirks are a likely explanation. Due to larger stdev (and two-band structure tracked as CSIT-1798) I did not do per-patch investigation there.

[7] https://csit.fd.io/trending/#eNrVUsGOwiAQ_ZruxUxSZsWePOj2PwzS0TYpLRlYtvr1S4kJNUazt40XYHgz7z1ecH5kOjjqt4XcF9W-wKpr4lJ87lZxC9YCDnAdEERZngmt0JOUGwXcGAXBHEI7Og-b9RGEBvIt9Djpo3KUCkxwYFHiGkQwEAnnBjDMswp-zSrNt7-TzIhtLxl5biQPKCa14Lr5y7gnt9D6s99McGJlyHVXyizx8RnXMc8MCX0v7S92gd5CqOrU8Sp8lOnJ0xQqUYIKp_-J_sHG2wcv649hZJN-fzy7dvwBz53qXbqKw6mYu2X9C0I46Ek

Original issue: https://jira.fd.io/browse/CSIT-1980

[vvalderrv]

Vhost test also show similar regression in trending [7], twice, also confirming compiler quirks are a likely explanation. Due to larger stdev (and two-band structure tracked as CSIT-1798) I did not do per-patch investigation there.

[7] https://csit.fd.io/trending/#eNrVUsGOwiAQ_ZruxUxSZsWePOj2PwzS0TYpLRlYtvr1S4kJNUazt40XYHgz7z1ecH5kOjjqt4XcF9W-wKpr4lJ87lZxC9YCDnAdEERZngmt0JOUGwXcGAXBHEI7Og-b9RGEBvIt9Djpo3KUCkxwYFHiGkQwEAnnBjDMswp-zSrNt7-TzIhtLxl5biQPKCa14Lr5y7gnt9D6s99McGJlyHVXyizx8RnXMc8MCX0v7S92gd5CqOrU8Sp8lOnJ0xQqUYIKp_-J_sHG2wcv649hZJN-fzy7dvwBz53qXbqKw6mYu2X9C0I46Ek