Benchmarking is an important tool to quickly be able to see the effect of your tuning efforts and also determine the limits of what your cluster is capable of. However, it's important that your benchmarks reflect the type of workload that you would be running normally on your Ceph cluster. It is pointless to tune your Ceph cluster to excel in large block sequential reads and writes if your final intention is to run highly latency sensitive OLTP databases on it. If possible, you should try and include some benchmarks that actually use the same software as your real-life workload. Again in the example of the OLTP database, look to see if there are benchmarks for your database software, which will give the most accurate results.