The more disk benchmarking I try to do, the more I want to more accurately simulate the actual workloads running in my system. I’m looking for a way that I can collect data about ZFS filesystem usage over a period of time to then craft fio benchmarks closer to what actually happens in the system.
I don’t need to get overly scientific - for the output I’m looking for something like:
- % mix of reads vs writes
- block size the application tried to write
- average queue depth (how ‘busy’ the system is)
Ideally I’d let this run for a few days to collect a sufficient sample.
Does something like this exist?