There's no single metric for 'faster', it depends on multiple factors (hw, type of workload, features used, ...). While there are fsync updates, the perf regression is in this case related to the update, quoting from the pull request use ticket space reservations for data, fair policy using the same infrastructure as metadata This is changing behaviour on a fundamental level, while still using a well tested infrastructure, there are different characteristics of data vs metadata. Such things may need time to fine tune, 5.10 is the first release where this got exposed to user testing, on hw and workloads that we haven't covered or noticed that the perf dropped. Testing focus is on reliability, performance comes next.