>
> I got some interesting results with this benchmark that I'd
> like some help interpreting. This test is on a Windoze XP
> box comparing:
>
> ora_odbc_driver (Oracle supplied(?) ODBC driver)
> ms_odbc_driver (Microsoft supplied ODBC driver)
> ora_dbd_driver (DBD::Oracle)
>
> Here is my test script:
[script snipped]
>
> Here are the results:
>
> Benchmark: timing 100 iterations of ms_odbc_driver,
> ora_dbd_driver, ora_odbc_dri ver...
> ms_odbc_driver: 35 wallclock secs ( 2.15 usr + 4.05 sys =
> 6.21 CPU) @ 16.11/s
> (n=100)
> ora_dbd_driver: 124 wallclock secs ( 1.44 usr + 2.89 sys =
> 4.34 CPU) @ 23.06/s
> (n=100)
> ora_odbc_driver: 262 wallclock secs ( 3.46 usr + 4.91 sys =
> 8.37 CPU) @ 11.94/ s (n=100)
> Rate ora_odbc_driver ms_odbc_driver ora_dbd_driver
> ora_odbc_driver 11.9/s -- -27% -48%
> ms_odbc_driver 16.3/s 37% -- -29%
> ora_dbd_driver 22.8/s 92% 40% --
>
> -----------------------------------
>
> Ok I'm having a brain-fart - can someone clue me in as to how
> to read this last section that compares the different methods
> using percentages? Is the 1st column saying that
> ms_odbc_driver is 37% faster than the ora_odbc_driver, and
> the ora_dbd_driver(DBD::Oracle) is 92% faster than the
> ora_odbc_driver?? The last 2 columns have me confused too.
> Any help is appreciated.
Ok -- First, let me comment on the "test". I understand that connection
times are important, but only testing drivers speed based upon
connection/disconnection time is probably not a good test and results may
vary wildly... I *do* have a hard time I believe that the Oracle ODBC
driver is *slower* that DBD::Oracle.
I updated your script, made it a bit more generic and added my ODBC drivers
in, I get strange results like yours above. From my results, I don't
believe the cmpthese function is correct -- and is sadly broken. Here's
what I show, using a local Oracle database, ACCESS, a remote SQLServer
(which always has a fast connection) and a sadly underperforming DB2 server
(remote also). The SQLServer and DB2 server are both on 100MBit switched
connections to my machine.
Benchmark: timing 100 iterations of DBD_ODBC_ACCESS, DBD_ODBC_DB2,
DBD_ODBC_MSORACLE, DBD_ODBC_ORACLE, DBD_ODBC_SQLSERVER, DBD_ORACLE...
DBD_ODBC_ACCESS: 7 wallclock secs ( 0.82 usr + 1.11 sys = 1.93 CPU) @
51.76/s (n=100)
DBD_ODBC_DB2: 64 wallclock secs ( 0.75 usr + 0.63 sys = 1.38 CPU) @
72.36/s (n=100)
DBD_ODBC_MSORACLE: 7 wallclock secs ( 1.54 usr + 1.15 sys = 2.69 CPU) @
37.12/s (n=100)
DBD_ODBC_ORACLE: 5 wallclock secs ( 1.82 usr + 1.26 sys = 3.08 CPU) @
32.41/s (n=100)
DBD_ODBC_SQLSERVER: 6 wallclock secs ( 1.04 usr + 0.90 sys = 1.94 CPU) @
51.49/s (n=100)
DBD_ORACLE: 4 wallclock secs ( 1.07 usr + 0.67 sys = 1.74 CPU) @ 57.37/s
(n=100)
Rate DBD_ODBC_ORACLE DBD_ODBC_MSORACLE DBD_ORACLE
DBD_ODBC_ACCESS DBD_ODBC_SQLSERVER DBD_ODBC_DB2
DBD_ODBC_ORACLE 32.1/s -- -27% -40% -51%
-53% -75%
DBD_ODBC_MSORACLE 43.8/s 37% -- -18% -34%
-35% -65%
DBD_ORACLE 53.7/s 67% 22% -- -19%
-21% -57%
DBD_ODBC_ACCESS 66.1/s 106% 51% 23% --
-3% -48%
DBD_ODBC_SQLSERVER 67.9/s 112% 55% 27% 3%
-- -46%
DBD_ODBC_DB2 126/s 293% 188% 135% 91%
86% --
>From the results you have and I have, cmpthese seems broken -- or is
*definitely* not correct on Win32 because it's using usr and sys time, not
wallclock time. (I guess that depends upon your perspective of "broken" --
maybe just "broken on Win32 would be more correct, until proven otherwise).
Also note, timethese is sadly reporting that I can perform 72.36 connections
per second using DB2, even though 100 connections took more than a one
minute...
Also, for reference, my script (which will be in the subversion repository
by the time you read this) is:
-----------
#!perl -w
use strict;
use DBI;
use Benchmark qw(timethese cmpthese timeit countit timestr);
my %conns = (
DBD_ORACLE => sub { doconnect("dbi:Oracle:URLWINLT"); },
DBD_ODBC_ORACLE => sub {
doconnect("dbi:ODBC:PERL_TEST_ORACLE"); },
DBD_ODBC_MSORACLE => sub {
doconnect("dbi:ODBC:PERL_TEST_MSORACLE"); },
DBD_ODBC_SQLSERVER => sub {
doconnect("dbi:ODBC:PERL_TEST_SQLSERVER"); },
DBD_ODBC_DB2 => sub { doconnect("dbi:ODBC:PERL_TEST_DB2"); },
DBD_ODBC_ACCESS => sub {
doconnect("dbi:ODBC:PERL_TEST_ACCESS"); },
);
sub doconnect ($) {
my $connstr = shift;
my $dbh = DBI->connect($connstr,
$ENV{DBI_USER},
$ENV{DBI_PASS},
{ RaiseError => 1, PrintError => 1 }
) || die "Can't connect with $connstr: $DBI::errstr";
}
timethese 100, \%conns;
cmpthese 100, \%conns;
---------
Regards,
Jeff
