.
In a recent tweet I wanted to highlight the importance of knowing what you measure with pgbench, because looking at “Transactions per second” without knowing if you are in shared buffer cache hits, or filesystem cache hit, or storage cache hit, or physical read… is just meaningless:
When you run the #postgres builtin pgbench, you should know:
– the size of the dataset on disk, about 15MB x scale
– the size of accessed shared_buffers (I see about 90% of size for a select-only workload)
– the size cached by the filesystem (looks like I have 2x here![]()
Recent comments
3 years 5 weeks ago
3 years 17 weeks ago
3 years 21 weeks ago
3 years 22 weeks ago
3 years 27 weeks ago
3 years 48 weeks ago
4 years 16 weeks ago
4 years 46 weeks ago
5 years 30 weeks ago
5 years 31 weeks ago