Re: [Gnumed-devel] Re: Document for introducing GnuMed (regarding India's project opportunity)

[Ian Haywood]

Karsten Hilbert wrote:

> No. We avoid them because they will not work without dblink.
> dblink works for separate databases no matter on which host
> each one is on.
Let's say you wanted to show a list of new path. results across
all patients, with the name and DOB of the patient.

At present, you would ask the 'clinical' server for a list of new results,
then for *each result* ask demographics for the name and address (by
instantiating cIdentity with the patient ID)
So 30 results = 30 little queries to demographics. (This is the speed penalty,
which, in my experience was so bad it rendered the client unusable, even on a 
local
database)

> > Perhaps we should reconsider requiring distribution, those few who want 
> > distributed databases [1]
> > can use dblink () and suffer the speed penalty currently enjoyed by 
> > everyone.
>
> Speed penalty ? And how would they use dblink lest it be
> integrated right from the beginning ?

An alternative is a view v_new_results. On most setups its an
ordinary view, and fast, as it provides all the info in one query,
with distrivuted databases it exits on (say) the clinical server
using dblink () to get the demographics stuff.
The client can't tell the difference (but the user will certainly notice,
unless they have a very fast server)

Ian

signature.asc
Description: OpenPGP digital signature