[Issue 10236] fragmentation makes mdb_page_alloc slow

https://bugs.openldap.org/show_bug.cgi?id=10236

--- Comment #6 from aalekseyev@janestreet.com --- I'm afraid I also found that fixing that bug erases a large part of the perf saving. A lot of saving came from dropping some "inconvenient" free list pages, thus accidentally defragmenting it.

So: Baseline: >3h (killed), 500 ms/brick towards the end With bug: ~2min, 23.4GB, 0.1 ms/brick towards the end Without bug: ~20min, 23.0GB, 30 ms/brick towards the end

So overall saving is "only" one order of magnitude, not two or more as I fooled myself into believing.