[ https://issues.apache.org/jira/browse/AVRO-4045 ]
David Mollitor deleted comment on AVRO-4045:
--------------------------------------
was (Author: belugabehr):
During some micro-benchmarking, I found that there was a significant overhead
to calling the JDK method Arrays#equals. For short strings, the difference in
performance was two orders of magnitude. I expected some overhead, but was
surprised by the final outcome.
However, for longer strings with long (e.g., 16+ characters) common prefixes
the vectorized performance was 50% better.
So, as a compromise, for short strings, use the existing method, for longer
strings, use the Vectorized methods.
> Use JDK to Compare Byte Array Lexicographically
> ------------------------------------------------
>
> Key: AVRO-4045
> URL: https://issues.apache.org/jira/browse/AVRO-4045
> Project: Apache Avro
> Issue Type: Improvement
> Components: java
> Affects Versions: 1.11.3
> Reporter: David Mollitor
> Assignee: David Mollitor
> Priority: Minor
> Labels: pull-request-available
> Time Spent: 40m
> Remaining Estimate: 0h
>
> The JDK introduced a new compare function for byte arrays in JDK9.
> Leverage this vectorized and "IntrinsicCandidate" (via ArraysSupport.java)
> implementation for Avro.
> [https://docs.oracle.com/en/java/javase/11/docs/api/java.base/java/util/Arrays.html#compare(byte%5B%5D,int,int,byte%5B%5D,int,int)]
> https://github.com/openjdk/jdk17/blob/4afbcaf55383ec2f5da53282a1547bac3d099e9d/src/java.base/share/classes/java/util/Arrays.java#L5804
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
