Re: Ada vs SQLite3 benchmark

["Dmitry A. Kazakov" <mailbox@dmitry-kazakov.de>]

On Thu, 18 Sep 2014 01:08:46 -0700 (PDT), briot.emmanuel@gmail.com wrote:

> Why did you run the benchmarks without optimization ? That seems
> inconsistent. If you are measuring performance, you should run with full
> optimization on I think.

Optimization could remove or rearrange parts of code which would not happen
in a real-life case. For example doing something like

   for I in 1..1000 loop
      N := I;
   end loop;

could be optimized to N := 1000.

IMO, not optimized code is a better measure for algorithmic complexity.
 
> Also, it would be interesting to use the following pragmas (combined or
> not) in sqlite, since they can impact performance significantly:
>      pragma journal_mode=WAL;
>      pragma synchronous=OFF;    (unless the Ada code is also running fsync() regularly)
>
> I think the latter in particular will significantly change the time measured for sqlite.

Thanks for pointing this.

Regarding Ada, it was strictly Ada.Direct_IO, nothing else. Ada.Direct_IO
does not have Flush [*]. As far as I can tell GNAT's implementation of
Ada.Direct_IO.Write is fwrite not followed by fsync. So forcing SQLite to
sync might be unfair. However, the intended use surely must sync upon
commit.

-----------
* Maybe it is worth an AI to add Flush to Direct_IO.

-- 
Regards,
Dmitry A. Kazakov
http://www.dmitry-kazakov.de