In 2306 cycle, VPP merged two [0] changes [1] with performance impact on ipsec tests. From cursory investigation, this seem to be a bigger regression and smaller progression, but different archs and scales can be impacted differently. For example on 3n-alt one must look at large scales [2] so the results are not affected by CSIT-1899, similarly for VPP-2072.

Review all ipsec tests to confirm there is no additional regression hiding in the trending.

[0] 38505: vppinfra: AES-CBC and AES-GCM refactor and optimizations | https://gerrit.fd.io/r/c/vpp/+/38505
[1] 38544: vppinfra: small improvement and polishing of AES GCM code | https://gerrit.fd.io/r/c/vpp/+/38544
[2] https://csit.fd.io/trending/#eNqVkUsOwjAMRE8TNshSUwhlw4LSe6CQmjaiHysxn3J6QkFKu2BBFo6UN_JMbM-9w6PHZidULrJcpJktQxGr_TJcNyJYdaAbhnVSYUry0WQygZLKCxg3EPcgldyeQBpAri2tLXk0MgmHu8bfITydtEewHYNGn6pNZVponXu7pIe3S3nlmWUkVA-R_AoS5dqhjvpPvkgZ_cTnj7SxxdnpFr19Yuwzfj8qTJjnBJq5PQ80od8hZIVQxaLrXTsuQRUvKnVzOQ