[
https://issues.apache.org/jira/browse/LUCENE-5109?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13774514#comment-13774514
]
Paul Elschot edited comment on LUCENE-5109 at 9/23/13 5:39 PM:
---------------------------------------------------------------
Patch of 23 september: as announced yesterday.
I tried benchmarking with index divisor 128 instead of 256. It is indeed a
little bit faster for far advanceTo operations.
I used this code snippet in the benchmark to avoid the EliasFanoDocIdSet being
used when it is not advisable:
{code}
new DocIdSetFactory() {
@Override
public DocIdSet copyOf(FixedBitSet set) throws IOException {
int numValues = set.cardinality();
int upperBound = set.prevSetBit(set.length() - 1);
if (EliasFanoDocIdSet.sufficientlySmallerThanBitSet(numValues,
upperBound)) {
final EliasFanoDocIdSet copy = new EliasFanoDocIdSet(numValues,
upperBound));
copy.encodeFromDisi(set.iterator());
return copy;
} else {
return set;
}
}
}
{code}
The sufficientlySmallerThanBitSet method currently checks for upperbound/7 >
numValues.
That used to be a division by 6, I added 1 because the index was added.
Anyway, "advisable" will depend on better benchmarking than I can do...
was (Author: [email protected]):
Patch of 23 september: as announced yesterday.
I tried benchmarking with index divisor 128 instead of 256. It is indeed a
little bit faster for far advanceTo operations.
I used this code snippet in the benchmark to avoid the EliasFanoDocIdSet being
used when it is not advisable:
{code}
new DocIdSetFactory() {
@Override
public DocIdSet copyOf(FixedBitSet set) throws IOException {
long numValues = set.cardinality();
long upperBound = set.prevSetBit(set.length() - 1);
if (EliasFanoDocIdSet.sufficientlySmallerThanBitSet(numValues,
upperBound)) {
final EliasFanoDocIdSet copy = new EliasFanoDocIdSet(numValues,
upperBound));
copy.encodeFromDisi(set.iterator());
return copy;
} else {
return set;
}
}
}
{code}
The sufficientlySmallerThanBitSet method currently checks for upperbound/7 >
numValues.
That used to be a division by 6, I added 1 because the index was added.
Anyway, "advisable" will depend on better benchmarking than I can do...
> EliasFano value index
> ---------------------
>
> Key: LUCENE-5109
> URL: https://issues.apache.org/jira/browse/LUCENE-5109
> Project: Lucene - Core
> Issue Type: Improvement
> Components: core/other
> Reporter: Paul Elschot
> Assignee: Adrien Grand
> Priority: Minor
> Attachments: LUCENE-5109.patch, LUCENE-5109.patch, LUCENE-5109.patch
>
>
> Index upper bits of Elias-Fano sequence.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
