On May 21, 5:24 pm, punkish <punk.k...@gmail.com> wrote: > Yeah, pretty good > > Benchmark: timing 10000 iterations of query_dbh, query_mem... > query_dbh: 5 wallclock secs ( 3.20 usr + 0.99 sys = 4.19 CPU) @ > 2386.63/s (n=10000) > query_mem: 6 wallclock secs ( 2.37 usr + 1.26 sys = 3.63 CPU) @ > 2754.82/s (n=10000)
Just for my sanity, can you do another pass where you're not compiling SQL in the cases where you have cache hits and don't need it?