John Sterrett is the latest smart person to take a look at automatic seeding of Availability Groups:
The 600 GB databases took about 66 minutes to seed across the network from a primary replica to the secondary replica. I noticed 1.4 Gbps of consistent throughput during the seeding process. This makes a lot of sense as it caps out around what the storage system can deliver in this environment.
The first thing I would look at for benchmarking throughput for network activity would be the bytes sent per second from the primary replica and bytes received per second on the secondary replicas.
John also includes an extended event session statement to track seeding. Great read.
Kevin,
Thanks for including me in your direct seeding post.
Regards,
John