Details
-
Improvement
-
Status: Open
-
Major
-
Resolution: Unresolved
-
10.6.1.0
-
None
-
unix
-
Urgent
Description
Using OFFSET and FETCH isn't any faster for paging than using existing JDBC methods:
//p is page number, n is rows per page
setMaxRows(n*p);
setFetchSize;
//...
rs.absolute((p-1)*n);
When used on a table with blobs, both the JDBC way, and the OFFSET/FETCH way are unexpectedly slow ('unexpectedly' because I am not even retrieving the BLOB column).
I need a way to do paging that doesn't slow down proportionately to the page I am requesting. It must also maintain this performance on tables with blobs. Following is discussion from users mail list, year or so ago.
> 2) what are the performance implications for users of the embedded
> driver? In particular, with the embedded driver I am hoping that
> this feature allows portions of a result set to be retrieved without
> the overhead of retrieving the entire result set. For example, if I
I am afraid that with embedded driver, you will only save a little CPU
(by avoiding some JDBC calls) since under the hood, the code siphons
off the rows till it hits the offset, so if you have a large offset,
you will still incur reading of those rows (modulo page caching). In
client/server driver context the savings are larger, of course, in
that fewer rows are sent over the wire. For simple queries that can
use an index, the optimizer could make use of the offset information
to avoid reading the entire row when skipping rows before offset, just
counting rows in the index to get to the first qualifying row, but
this optimization is not yet implemented.
Often, this feature is used together with ORDER BY which would entail
some sorting of the result set and then all the rows would have to be
read anyway. Again, for some simple queries, sort avoidance is used by
the optimizer, so optimization is still possible for for such queries.
If you think this optimization is an important capability feel free to
file an improvement issue for it.
Attachments
Issue Links
- is related to
-
DERBY-6335 Improve performance of FETCH NEXT/OFFSET with ORDER BY
- Open
- relates to
-
DERBY-1506 full table scans of tables which don't use indexes, which have blobs, but don't reference blob data still read all pages of the table
- Open