On Fri, 2009-04-03 at 18:03 +0100, Greg Stark wrote:
> I wonder if we need a whole class of index algorithms to deal
> specifically with read-only tables
I think we can drop the word "index" from the sentence as well.
"Read-only" isn't an isolated case. Often you find many read-only tables
alongside rapidly changing tables. So even the busiest of databases can
benefit from read-only optimisations. So I want MVCC *and* read only,
not MVCC everywhere (or MVCC nowhere if customer changes horses to get
read-only benefits elsewhere).
Having changes to those tables cause much heavier additional work is OK,
if judged on a cost/benefit basis. So the case I care about ought to be
called "read-mostly" but we're talking write:read ratios of millions:1.
-- Simon Riggs www.2ndQuadrant.comPostgreSQL Training, Services and Support
