Keywords: DataAccWG
Hi,
Things got a little better after I assumed tai column and an index
in DIASource. Now the disk io highly depends on selectivity of
the cone. If I use 5% for both the low-volume query "Select time series
data for given cone", and the high-volume "Find all transients
in given epoch", I am getting a pretty reasonable 625K disk heads.
If I change the selectivity to 50% (to follow Kem's suggestion),
it is going up to 2.2 million
That is because:
SELECT o.objectId, o.ra, o.dec, d.TAI, d.*
FROM Object o, DIASource d
WHERE o.ra BETWEEN ra1 + cos(decl1)*(0.5)
AND ra1 - cos(decl1)*(0.5)
AND o.decl between decl1 +0.5 and decl1 - 0.5
AND o.objectId = d.objectId
ORDER BY o.objectId, d.TAI ASC
For 50% cone selectivity we select 5 million Stars (from a single
partition), after join with DIASource it becomes 25 million
(5 DIASource per star). Now we have to join the result with
the TAI index which is big because we have 45 billion entries
in the DIASource table. And finally, we have to fetch DIASource
row (all 25 millions).
cheers,
Jacek
_______________________________________________
LSST-data mailing list
[email protected]
http://www.lsstmail.org/mailman/listinfo/lsst-data
