Jim Nasby <jim@nasby.net> writes:
> Yeah, but the problem we run into is that with every backend trying to run the clock on it's own we end up with high
contentionagain... it's just in a different place than when we had a true LRU. The clock sweep might be cheaper than
thelinked list was, but it's still awfully expensive. I believe our best bet is to have a free list that is actually
usefulin normal operations, and then optimize the cost of pulling buffers out of that list as much as possible (and let
thebgwriter deal with keeping enough pages in that list to satisfy demand).
Well, maybe, but I think the historical evidence suggests that that
approach will be a loser, simply because no matter how cheap, the
freelist will remain a centralized and heavily contended data structure.
IMO we need to be looking for a mechanism that has no single point of
contention, and modifying the clock sweep rules looks like the best way
to get there.
Still, he who wants to do the work can try whatever approach he feels
like.
regards, tom lane