14 lines
632 B
Plaintext
14 lines
632 B
Plaintext
|
Indexing
|
||
|
|
========
|
||
|
|
|
||
|
|
The indexing omits the table BY_SPO, indexing on all three arguments but
|
||
|
|
uses BY_SP instead. It was thought not to gain that much. This
|
||
|
|
assumption was verified on real data (100,000 triple). Introducing
|
||
|
|
BY_SPO gave a slowdown of about 10%. Appearently the relatively high cost
|
||
|
|
for indexing on the complicated object field outweights the benefits.
|
||
|
|
|
||
|
|
The patch is left for reference in BY_SPO.diff. Note that after applying
|
||
|
|
this patch the test cases no longer pass. As the failing cases were
|
||
|
|
irrelevant to the performance test I never bothered fixing this due to
|
||
|
|
the failed performance test.
|