search

dbgroup-at-ucsc / dbtune

This research project aims to develop tools in order to make index tuning easier and more effective.
http://users.soe.ucsc.edu/~alkis/tuning/
Other
5 stars 3 forks source link

Vote up not updating scores correctly #298

Closed ivotron closed 12 years ago

ivotron commented 12 years ago

Not sure if this is the way WFIT should behave. I need to investigate and run Karl's code on this to see the output, but on the DBTune side, the following doesn't work:

-- ########################
-- GOOD positive vote
-- ########################
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
UPDATE one_table.tbl set a = a+1 WHERE a BETWEEN        0 AND  500000;
UPDATE one_table.tbl set a = a+2 WHERE a BETWEEN  500000 AND  900000;
UPDATE one_table.tbl set a = a+3 WHERE a BETWEEN  9000000 AND 11000000;
-- vote A up so that it doesn't get dropped
UPDATE one_table.tbl set a = a+4 WHERE a BETWEEN 11000000 AND 15000000; <--- this DO drops the index from the recommendation
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
SELECT a FROM one_table.tbl WHERE a = 2;
ivotron commented 12 years ago

@npolyzotis , do you have any ideas why this might be happening? Is there any difference between the way the score is updated for a positive and negative vote?

I'll keep looking at the code to see if I can figure it out

npolyzotis commented 12 years ago

Something is going wrong with the voting. Is A the only index in your candidate set? If so, then you can check the work function values for the two states (A materialized vs no) right after the vote. They should indicate that A is the best state. You can also check the same experiment but casting the vote earlier--it might just be that the vote comes too late.

On May 8, 2012, at 21:16, Ivo Jimenezreply@reply.github.com wrote:

@npolyzotis , do you have any ideas why this might be happening? Is there any difference between the way the score is updated for a positive and negative vote?

I'll keep looking at the code to see if I can figure it out

Reply to this email directly or view it on GitHub: https://github.com/dbgroup-at-ucsc/dbtune/issues/298#issuecomment-5592489