Has anyone ever tried using in-situ OAM for testing latency? It would seem to me that the protocol is ideal for this purpose: it adds metadata to actual user packets. The Linux kernel has support for it as of 5.15, Cisco and Juniper have implemented it as well.

Here’s an article on the Linux kernel implementation: https://legacy.netdevconf.info/0x14/pub/papers/32/0x14-paper32-talk-paper.pdf

If anyone has tried this, how did it turn out?