16 mutext lock/unlocks per query?

[Paul A Vixie]

> Which platforms are considered as having lightweight mutexes?

bsd is what we had in our hands at the time we did the design.

it simply never occurred to us that there would ever be a thread
implementation without inline functions and without usermode TSI
before syscall.

that doesn't make us right, but it should explain why 15 mutexes
per transaction didn't seem like the worst thing in the world at
the time we did the design.

i suspect that a "big lock" design would be faster for most work
loads.  or at least a bigger lock granularity than we have now.
mutexes are a LOT more expensive on non-bsd systems than we ever
expected.