+chongz@

This is a performance regression, so P1. Understanding the impact is P1, but let's reassess priority from there.

So the initial bump looks to be adoption. The data around the Nov 6,7 shows about 5ms of latency 18ms->23ms.

chongz@ this is similar numbers you saw on the perf waterfall right?

I believe the 18ms->23ms regression is what we are expecting from Vsync Aligned Input.

The 18ms->23ms = +27.8% regression is similar to the one we saw in Finch Trial mentioned in #0.
e.g. A +38.2% sum regression in 50th percentile where the count is roughly the same (Δ: -5.9%).

This 5ms regression sounds reasonable to me since Vsync Aligned Input could cause as much as half a frame delay in average.

The Speed Releasing team is trying to be a bit more rigorous about justifying regressions.

Do we have a doc or something we can point to when claiming that we should accept this?

Re #6: No I don't think there is a doc or something we can point to... IIRC we discussed offline and agreed that the additional half-frame latency is unavoidable if we want to align events...

Yup, we've had this discussion as the input team. We want something we can point folks in general to which justifies this regression.

High speed camera tests were conducted and recorded here:

https://docs.google.com/spreadsheets/d/1wgN2sbqzefWXRqeIqOHy-xUC6R5ijsLisOgqNduoXzQ/edit#gid=544614617

Thanks dave for attaching the doc. Looking at the doc and the UMA it doesn't seem to be much of a regression beyond what we expected. Closing this for now but let me know if anyone has any comments and wants further investigation.