On the recurring debate of threading vs. forking, I was giving it a fwe
thoughts a few days ago, particularly with concern to Linux's memory model.
On IA32 platforms with over 4 gigs of memory, any one process can only
"see" up to 3 or 4 gigs of that. Having each postmaster fork off as a new
process obviously would allow a person to utilize very copious quantities of
memory, assuming that (a) they were dealing with concurrent PG sessions, and
(b) PG had reason to use the memory.
I'm not entirely clear on threading in Linux - would it provide the same
benefits, or would it suddenly lock you into a 3-gig memory space?
steve
