Some performance numbers from a simple test

[Thomas Sødring]

On 02/03/2017 04:43 PM, Petter Reinholdtsen wrote:
> I wrote a simple script to import a short post journal I had laying
> around, and did some simple measurements.  I ran top and wrote down VIRT
> and RES values just after starting, and after the import was done.  The
> import script loaded 8333 journal entries with case (sak) = mappe, empty
> record/registrering and document title = dokumentbeskrivelse.
>
> The import took 3 minutes 35 seconds (ie 1/39 seconds per insert) and
> increased VIRT size by 281036 kilobytes (ie 34 kibibytes per insert).
> RES increased by 392454 kilobytes (ie 48 kibibytes per insert).
That sounds awful! I'll try and do something similar either over the
weekend or early next week! I'll also start the core connected to a
memory profiler to see what's happening!
> I hope I am doing it wrong, as this was quite a lot more memory usage
> than expected. :)
I doubt your doing it wrong! Quite a lot more is an understatement!! The
interface is that simple! My simple guess at the moment is that
hibernate somehow is loading the entire data structure (tree) in memory.
Hibernate is great if it's configured properly! If it's not then it can
be a nightmare!
>
> And using this naive approach will take 13 years to insert 20 million
> entries. :)
>
> Btw, is searching implemented?
Partly in the 0.2 branch. It's persisting to elasticsearch. I'll try and
commit something working over the weekend if you want to try search via
elasticsearch!

"Official" search is via OData, that's a whole other story!

 - Tom