>
> * Consider padding the spinlock structure to have spinlock registers
>   on different CPU cache line
>
>   http://archives.postgresql.org/pgsql-hackers/2005-10/msg00498.php