Kynn Jones wrote:
> I have large database that needs to be built from
> scratch roughly once every month. I use a Perl script to do this.
>
> The tables are very large, so I avoid as much as possible using
> in-memory data structures, and instead I rely heavily on temporary
> flat files.
>
> I have solved this general problem in various ways, all of them
> unwieldy (in the latest version, the script generates the serial ids
> and uses Perl's so-called "tied hashes" to retrieve them when needed).
>
> TIA!
>
> kj
I have done this exact same thing. I started with tied hashes, and even
tried BerkeleyDB. They only helped up to a point, where they got so big
(a couple gig if I recall correctly) they actually slowed things down.
In the end I used a stored proc to do the lookup and insert. In the
beginning its not as fast, but by the time the db hits 20 gig its still
going strong, where my BerkeleyDB was becoming painful slow. (I
recently thought of trying a sqlite table, I've had good luck with them,
they can get pretty big and still be very fast... but never got around
to trying it.)
So... not really an answer (other than I used a stored proc), but I'd be
interested in alternatives too.
-Andy
