Re: Auto-tuning work_mem and maintenance_work_mem

On Thu, Oct 10, 2013 at 6:27 PM, Josh Berkus <josh(at)agliodbs(dot)com> wrote:
>> More generally, Josh has made repeated comments that various proposed
>> value/formulas for work_mem are too low, but obviously the people who
>> suggested them didn't think so. So I'm a bit concerned that we don't
>> all agree on what the end goal of this activity looks like.
>
> The counter-proposal to "auto-tuning" is just to raise the default for
> work_mem to 4MB or 8MB. Given that Bruce's current formula sets it at
> 6MB for a server with 8GB RAM, I don't really see the benefit of going
> to a whole lot of code and formulas in order to end up at a figure only
> incrementally different from a new static default.

Agreed. But what do you think the value SHOULD be on such a system?

I suggest that it's pretty reasonable to assume that even a
developer's personal machine will likely have 8GB or so by the time
PostgreSQL comes out, so tuning work_mem on that basis is not
unreasonable. Plus, even if it has less, a developer probably won't
have 100 connections.

I guess the point I'm making here is that raising the default value is
not mutually exclusive with auto-tuning. We could quadruple the
current defaults for work_mem and maintenance_work_mem and be better
off right now, today. Then, we could improve things further in the
future if and when we agree on an approach to auto-tuning. And people
who don't use the auto-tuning will still have a better default.

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