I tested for range queries times on the DB using the DATA table (no BLOBs).
Problems related to DB size and time required to populate it still need to be solved. I haven't thought about it yet. I guess that populating would be quicker using C-language interface, but I have no clue about how to deal with the size! Probably the only solution is to move to BLOBs. Any ideas?
Here below you can find the times required to perform the same range queries on both the mzRTreeDB and the old published mzRTree on the same data file. File description can be found in the mzRTree paper.
Results have been evaluated in Java using Xerial jdbc (www.xerial.org/trac/Xerial/wiki/SQLiteJDBC
) and the same benchmark class test used for mzRTree published results.
Query results from the DB are left as they are, no manipulation is counted for in time estimations. Otherwise, accessing the query results under Java would be extremely time wasting. I assume using C-language interface everything should work faster. Maybe Matt could test for it.
Results are quite good, just to have an idea:
(Rt here stands for Scan)
% mzRTreeDB performance (Windows new jdk: 1.6)
Time "Load" mzRTreeDB = 6.9493E-5
Time mzRTree x10 (20 Rt-Spectra) = 0.09737985
Time mzRTree x10 (5Da-Chroms) = 0.341934281
Time mzRTree x10 (60Rtx5Da-Chroms) = 0.388018744
Time mzRTree x10 (200Rtx5Da-Chroms) = 0.399810525
% mzRTRee published results on the same data file (EXP1 from paper) (Linux old jdk: 1.5)
Time Load mzRTree = 0.01
mzRTree x10 (20 Rt-Spectra) = 0.34
Time mzRTree x10 (5Da-Chroms) = 0.77
Time mzRTree x10 (60Rtx5Da-Chroms) = 0.01
Time mzRTree x10 (200Rtx5Da-Chroms) = 0.03
% mzRTRee last results on the same data file (EXP1 from paper) (Windows new jdk: 1.6)
Time Load mzRTree = 0.013071723
Time mzRTree x10 (20 Rt-Spectra) = 0.602019962
Time mzRTree x10 (5Da-Chroms) = 0.3692395
Time mzRTree x10 (60Rtx5Da-Chroms) = 0.012800634
Time mzRTree x10 (200Rtx5Da-Chroms) = 0.032766208
Don't worry about the "switch" between the chroms and spectra access times for mzRTree. mzRTree underwent some minor changes since the publication.
What to do next?
Sara Nasso, Ph.D. student
Department of Information Engineering (DEI)
University of Padova
Via Ognissanti 72 35129 PADOVA, ITALY