Re: Thinking about breaking up the BufMgrLock

Greg Stark <gsstark(at)mit(dot)edu> writes:
> I recall the clock sweep having
> been recommended in class as having most of the best properties of LRU with
> very low cost in the critical path.

Right. The "pending move to front" idea that I suggested is basically a
variant of a clock algorithm: it takes two trips through the LRU list
before a page falls off and becomes eligible for replacement. (It's
even closer to the "second chance" clock algorithm.)

The $64 question with any of these things is how much performance at the
cache-management level are we willing to give up in order to gain
low-level efficiency? We probably don't want to go very far in that
direction. But maybe a clock scan will be a good compromise.

regards, tom lane