I was asked to look at performance degradation of a specific query after a system had upgraded to SQL Server 2016. The performance degradation was mitigated by changing database compatibility level from 2016 to 2012. Lots of stuff could be happening there... I'll hopefully address it at a later time. :-) Before looking at the individual query, I wanted to check overall behavior of the system. Tuning an individual query without checking for overall system health can be a very frustrating experience.
The backup behavior on the instance was concerning - another item I'll have to address another day. While speculating what could be leading to database backup misbehavior on the system, I had a lot of trouble reconciling the read IO numbers reported by the Resource Governor resource pool, the perfmon LogicalDisk counters, and the perfmon Backup/Recovery counters. It seemed like Resource Governor was reporting twice as many read operations and twice as many read bytes for the backup operation as perfmon LogicalDisk Counters and the Backup/Recovery counters.
Here's a test script I used to explore this with a test database on my laptop, confirming the results on a few VMs as well.
And... here are the results from my test database. Not sure why this happens. I'll create a feedback item for it - I'd like to rely on the numbers that come from Resource Governor. Especially for backup activity - having the ability to separate backup activity into its own resource pool and account for its reads, read bytes, and read service time separately could be very powerful. But its gotta tell me the truth first :-)
If you really, really love this stuff you can see how I started wading into these waters - complete with some perfmon graphs, in the thread following the tweet linked below :_)