Bug#317518: New version of libsql-statement-perl is very, very much slower

[Paul Beardsell]

Tyni,

It gets worse and worse.  My "standard" job run for the purpose of
libsql-statement-perl benchmarking:

Box 1: Woody with version 0.1020-2:  173.10s real   124.18s user    
0.55s system

Box 2: Sarge with version 1.11:  852.88s real   849.86s user     1.55s system

Box 2: Sarge with version 1.14-1: 896.71s real   893.45s user     1.55s system

Notes:  The Woody and Sarge boxes are similar but not exactly the
same.  The Woody box is live and busy.  The Sarge box is otherwise
idle.  So I cannot directly and exactly compare Woody
lib-sql-statement with Sarge except to say that the new pure Perl
version is MUCH SLOWER.  Also, the purported performance improvements
are not there in the 1.11 to 1.14-1 comparison - the package has got
slower, not faster.  This last comparison I have repeated to make sure
I did not make a mistake.  Note that Box 2 remained identical between
the tests except for the libsql-statement-perl version change.

Thanks for looking at this and for all your work generally BUT I am
facing a forced move to SQLite - which I do not want to do!  My
preference is to stay with CSV text databases and
libsql-statement-perl.  Can we not have a libsql-statement-perl-fast
package (which is just the old Woody version) in Sarge?

My simple performance tests could easily be duplicated by those
claiming performance improvements!

Regards,
Paul

On 12/15/05, Niko Tyni <ntyni at iki.fi> wrote:
> On Sat, Jul 09, 2005 at 12:00:27PM +0100, Paul Beardsell wrote:
> > Package: libsql-statement-perl
> > Version: 1.11-1
> > Severity: normal
>
> > I depend upon libsql-statement-perl to manage small
> > plain text databases of approx 2000 rows.  Performance
> > was fine in the version of libsql-statement-perl
> > which came with woody but the newer version in sarge
> > is very, very much slower.
>
> Hi,
>
> could you please check if the current version in testing/unstable
> (1.14-1) fixes this? The upstream changelog mentions several
> optimizations since 1.11.
>
> Thanks,
> --
> Niko Tyni       ntyni at iki.fi
>


--
Paul Beardsell
Paul at Beardsell.com
http://beardsell.com/