On Thu, Jan 5, 2012 at 6:10 PM, Lee Hachadoorian
<lee.hachadoorian@gmail.com> wrote:
>
> Many of the smaller geographies, e.g. census tracts, do in fact have data
> for the vast majority of the columns. I am trying to combine it all into one
> table to avoid the slowness of multiple JOINs (even though in practice I'm
> never joining all the tables at once). EAV sounds correct in terms of
> normalization, but isn't it usually better performance-wise to store
> write-once/read-many data in a denormalized (i.e. flattened) fashion? One of
> these days I'll have to try to benchmark some different approaches, but for
> now planning on using array columns, with each "sequence" (in the Census
> sense, not the Postgres sense) of 200+ variables in its own array rather
> than its own table.
Are you using arrays or hstore?