Thanks Chidera for your response.
Sorry for the confusion.
Currently we have two types of http requests that are doing the experiment. These are uploading and downloading files via http requests.
For upload, we're seeing higher h3 usage, ~70% of requests, but for some file types are lower.
For download, overall we're seeing only ~50% of requests are h3, which is way lower than our expectations.
If it's not h3, we use h1. The problem is Cronet's h1 is much slower than our current network stack (even though we're using h1, we do have some tunings to make it faster) which is affecting our overall latency metrics. Overall we're seeing 10% of latency regression for download side, and data suggests the main contribution is almost 50% of h1 usage, similar regression on upload side.
Our main blockers for using Cronet is the low usage of h3. We'd expect generally the h3 usage should be around 90% in hope that it would offsite some of the regression that is brought by h1 requests.