Jim Kellerman (JIRA) wrote:
[ https://issues.apache.org/jira/browse/HADOOP-2365?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#action_12549100 ]
Jim Kellerman commented on HADOOP-2365:
---------------------------------------
-1 on patch. This section of code should read:
{code}
int[] result = new int[nbHash];
for (int i = 0, initval = 0; i < nbHash; i++) {
initval = result[i] = Math.abs(JenkinsHash.hash(b, initval)) % maxValue;
}
return result;
{code}
Yes, this works too - it shouldn't matter in this specific case.
Jenkins' hash has very good avalanche behavior, so even 1 bit difference
in the initvalue yields a completely different hash.
However, thanks for finding my stupid mistake.
You're welcome. I'm using this class in a different Hadoop application,
where the problem became immediately apparent when I switched from my
home-grown BloomFilter implementation to this one.
--
Best regards,
Andrzej Bialecki <><
___. ___ ___ ___ _ _ __________________________________
[__ || __|__/|__||\/| Information Retrieval, Semantic Web
___|||__|| \| || | Embedded Unix, System Integration
http://www.sigram.com Contact: info at sigram dot com
