$dbh->prepare() and execute() take too long?

[jbrugger]

Oh wise Perl magicians

I have a query that looks something like this :

select ra.value, ra.ra_id, q.id as q_id, q.qid , q.parent, q.q, q.type
+, p.qid as pqid, if($ifstatement,'1','0') as selection , r.rpid from 
+r, ra, q left join q p on q.parent = p.id where r.id = ra.ra_id and r
+.q in (3,4,5) and ra.q = q.id order by ra.ra_id ASC,selection desc;

The 'explain' shows a good and nice run (no full table scans), and in mySQL it takes about 3 seconds to run (about 130.000 records) using  mysql -u user -p database < q.sql > output.txt
When doing this from whithin Perl however,

my $bm=  DEBUG_Benchmark->new();
$bm->bmStart();

$sth = $dbh->prepare("select ra.value, ra.ra_id, q.id as q_id, q.qid ,
+ q.parent, q.q, q.type, p.qid as pqid, if($ifstatement,'1','0') as se
+lection , r.rpid from r, ra, q left join q p on q.parent = p.id where
+ r.id = ra.ra_id and r.q in (3,4,5) and ra.q = q.id order by ra.ra_id
+ ASC,selection desc");

$sth->execute();

$bm->bmStop();
print $bm->getBmResult();

It takes about 12 seconds to run (Only the prepare and the execute, i did not even loop over the resultset).
Any idea on how to speed this up? Whit a smaller resultset, it takes about .003 seconds to prepare / execute.
I know DBI should be able to do mutch better, any ideas on how i can get a better result?
Thanks.

ps.
I run this on an UML (user mode linux), so there is a lot of overhead in the disk io, however, the same problem occurs on a 'normal' debian distribution, the difference is big. (2 sec - 8 sec)

ps.ps.
The resultset of the query looks like this:

value    ra_id    q_id    qid    parent    q(4,5,6)type    pqid    sel
+ection    rpid
1    5501    1918917    34    0    4    6    NULL    1    ed94d62745f5
+2f83cc6225c46cd1f012
1    5501    1918918    33    0    4    6    NULL    1    ed94d62745f5
+2f83cc6225c46cd1f012
4    5501    1918915    32    0    4    6    NULL    1    ed94d62745f5
+2f83cc6225c46cd1f012
2    5501    1918940    31    0    4    6    NULL    1    ed94d62745f5
+2f83cc6225c46cd1f012
6    5501    1918916    30    0    4    6    NULL    1    ed94d62745f5
+2f83cc6225c46cd1f012
...

*** UPDATE ***
Ah FOUND IT!!! Here is the answer!!!

Actually it IS the overhead from the UML! Time to step over to XEN.
After testing we found out that using apache ont the uml in a simplefied script still executed in 12 seconds, and on our life server in the same time as the commandline options (3 seconds).
ergo:

uml:
Command line: *** Benchmark *** : Execute: 4.636518 seconds
From Apache : *** Benchmark *** : Execute: 12.584603 seconds

life:
Command line: *** Benchmark *** : Execute: 3.079879 seconds
From Apache : *** Benchmark *** : Execute: 3.066677 seconds

Sorry to have bothered you all, it seems that apache uses the disk even more, cousing even more io overhead.