SDO_RELATE - EQUAL MASK - Different results on the same query

Question

Hello.
I'm trying to find all spatial duplication in the point layer (using certain tolerance). That layer has q-tree spatial index and appropriate tolerance in the user_sdo_geom_metadata table (let's say 1 meter). I have SDO_VERSION=9.2.0.5.0.
To perform that I use SDO_RELATE operator with EQUAL mask (I know that actually I can use other operators but lets consider EQUAL):
SELECT /*+ ordered */t1.OBJECTID, /*+ ordered */t2.OBJECTID from points_table t1, points_table t2 WHERE t1.OBJECTID < t2.OBJECTID
AND sdo_relate (t2.shape, t1.shape, 'mask=EQUAL querytype = JOIN') = 'TRUE'
The problem is:
If I create Q-tree index using SDO_LEVEL=8 query takes 10 minutes and returns lets say 1000 dup events
If I create Q-tree index using SDO_LEVEL=10 query takes 20 seconds and returns lets say 996 dup events
I'm confusing about that - Why the results are depend from SDO_LEVEL? For me it is more important then performanse.
Those missed duplication events are really duplication events (point closer to each other than 1 meter) and SDO_DISTANSE for those particlar poins returns 0. It is correct, but SDO_RELATE monitors that those points are not equal.
So how can I rely on the SDO_RELATE function? What SDO_LEVEL need I use to get correct results? I choosed Q-tree indexes because of performance I didn't get success to get result using R-tree indexes on big datasets at all.
Regarding polylines:
I suppose that I can get similar problem and for polylines (I’m also using same query to find dups polylines).
Sometimes SDO_RELATE finds polylines which ones are not really equal in the specified tolerance (1 meter). All cases that I saw – it was simple segments which ones share only one vertex but the other ends are spatially disjoint more that 3 meters. Why SDO_RELATE finds such cases as EQUAL?
Thank you very much in advance.

Answer 1

Thank you Siva for advice.
I've tried R-tree indexes and got reasonable performance on the tables with ~20k rows.
But when I try to ran that query on the table (points) with 250k rows it takes forever :(.
1) Create index:
Create index nodes_idx on nodes(SHAPE) indextype is mdsys.spatial_index parameters ('LAYER_GTYPE=POINT initial=10m next=5m sdo_rtr_pctfree=10 pctincrease=0');
/* actually 'initial=10m next=5m sdo_rtr_pctfree=10 pctincrease=0' options only decreased slightly performance */
2) Execute query:
SELECT /*+ ordered */t1.OBJECTID, /*+ ordered */t2.OBJECTID
from nodes t1, nodes t2 WHERE t1.OBJECTID < t2.OBJECTID AND
sdo_relate (t2.shape, t1.shape, 'mask=EQUAL querytype = WINDOW') = 'TRUE'
Plan Explanation:
| Id | Operation | Name | Rows | Bytes | Cost |
| 0 | SELECT STATEMENT | | 150M | 1095G| 505M|
| 1 | NESTED LOOPS | | 150M| 1095G| 505M |
| 2 | TABLE ACCESS FULL | NODES | 548K | 2043M| 411 |
| 3 | TABLE ACCESS BY INDEX ROWID| NODES | 274 | 1047K| 505M|
| 4 | DOMAIN INDEX | NODES_IDX | 27416 | | |
As far as I understand, oracle reserved for that query 1095G of memory?! Is it really necessary for that simple query and how can it be possible?
Regards, Denis.
Message was edited by:
Tenek