Re: Page replacement algorithm in buffer cache

On Tue, Apr 2, 2013 at 11:32 AM, Merlin Moncure <mmoncure(at)gmail(dot)com> wrote:
> That's a very fair point, although not being able to evict pinned
> buffers is a highly mitigating aspect. Also CLOG is a different beast
> entirely -- it's much more dense (2 bits!) vs a tuple so a single page
> can a lot of high priority things. But you could be right anyways.
>
> Given that, I wouldn't feel very comfortable with forced eviction
> without knowing for sure high priority buffers were immune from that.
> Your nailing idea is maybe the ideal solution. Messing around with
> the usage_count mechanic is tempting (like raising the cap and making
> the sweeper more aggressive as it iterates), but probably really
> difficult to get right, and, hopefully, ultimately moot.

One thought I had for fiddling with usage_count is to make it grow
additively (x = x + 1) and decay exponentially (x = x >> 1). I'm not
sure the idea is any good, but one problem with the current system is
that it's pretty trivial for a buffer to accumulate five touches, and
after that we lose all memory of what the frequency of access is, so a
pages of varying different levels of "hotness" can all have usage
count 5. This might allow a little more refinement without letting
the time to degrade the usage count get out of control.

But, having said that, I still think the best idea is what Andres
proposed, which pretty much matches my own thoughts: the bgwriter
needs to populate the free list, so that buffer allocations don't have
to wait for linear scans of the buffer array. That's just plain too
slow.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company