[Monetdb-developers] Single column select vs aggregation
Stefan de Konink
stefan at konink.de
Thu Feb 18 19:05:47 CET 2010
Op 18-02-10 18:23, Stefan Manegold schreef:
> On Thu, Feb 18, 2010 at 05:46:31PM +0100, Stefan de Konink wrote:
>> Attached is a trace of a single column select operation, with a max,
>> avg and equality.
>> The equality takes about 4 times more time than the avg operation.
>> From the plan I see that the operation is executed paralel, but I do
>> wonder why suddenly the uselect takes such a high amount of time.
> Which version of MonetDB?
CVS HEAD, 2 am.
> How configured / compiled?
> How was the mserver5 started (which command line options were given)?
--dbname=kvk --dbinit="include sphinx; include sql;" --set mapi_port=50001
> Any changes compared to the default monetdb5.conf?
No, not changed anything in there.
> Is the "kvk" column sorted?
No, it does include a partial sort (msb are sorted). A 32bit column
(kvks) that is (or either should be...) ends up with between 170ms ~ 220ms.
> Could you run the select query in the MAL debugger
> (DEBUG select kvk from kvk where kvk = 412657690010;),
> single step through it,
> set the debug mask to 2097152
> (debug 2097152)
> just before the algebra.uselect() is executed,
> and report the output on the server console after the algebra.uselect() has
> been executed?
Do you mean this?
mdb># _44 :=
#BAT_select_(b=tmp_5710): sampling: tmp1 = BATslice(b=tmp_5710,
#BAT_select_(b=tmp_5710): sampling: tmp2 = BAT_select_(tmp1=tmp_5704,
tl, th, tail);
#BAT_select_(b=tmp_5704): BAT_scanselect(b=tmp_5704, bn=tmp_5703, tl,
th, equi=1, nequi=0, lval=1, hval=1, nocheck=1);
#seqscan_eq_lng_void_tloc_oid_vid_nocheck_noinc[if ( simple_EQ(tl
bn=tmp_5703, tl, th, oid_ctr=598266, str_idx=0);
#BAT_select_(b=tmp_5704): tmp_5703: hkey=1, tkey=0,
#BAT_select_(b=tmp_5710): BAT_hashselect(b=tmp_5710, bn=tmp_5703, tl);
(building hash-table on the fly)
#BAThash: create hash(1196532);
#BAT_select_(b=tmp_5710): tmp_5703: hkey=1, tkey=0,
More information about the developers-list