>If you're not using a connection pool of some kind then you might as
>well forget query plan caching, because your connect overhead will swamp
>the planning cost. This does not mean you have to use something like
>pgpool (which makes some rather questionable claims IMO); any decent web
>application language/environment will support connection pooling.
>
>
Hmm, a question of definition - there's a difference between a pool and
a persistent connection. Pretty much all web apps have one connection
per process, which is persistent (i.e. not dropped and remade for each
request), but not shared between processes, therefore not pooled.