[
https://issues.apache.org/jira/browse/SOLR-15559?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17386540#comment-17386540
]
Gus Heck commented on SOLR-15559:
---------------------------------
*Default memory*
{code:java}
Benchmark Mode Cnt Score Error Units
MyBenchmark.testCharBuff thrpt 25 346896.291 ± 6717.362 ops/s
MyBenchmark.testClassic thrpt 25 204691.222 ± 2944.000 ops/s
{code}
speedup = ~2 microseconds, 40%
*Memory constrained to 8mb*
{code:java}
Benchmark Mode Cnt Score Error Units
MyBenchmark.testCharBuff thrpt 25 187599.422 ± 17714.713 ops/s
MyBenchmark.testClassic thrpt 25 127792.646 ± 8938.638 ops/s
{code}
speedup = 2.5 microseconds, 31%
*Memory constrained to 5mb*
{code:java}
Benchmark Mode Cnt Score Error Units
MyBenchmark.testCharBuff thrpt 25 209509.435 ± 11178.546 ops/s
MyBenchmark.testClassic thrpt 25 117720.233 ± 10968.465 ops/s
{code}
speedup = 3.7 microseconds, 43%
5mb was the lowes memory I could run the test under, I think there both tests
are regularly GCing, and in the unconstrained memory which should be several
GB, I suspect there was little GC. I did not verify this with a profiler
however. My conclusion s that this likely does some small good and no harm.
> Avoid object creation in splitSmart()
> -------------------------------------
>
> Key: SOLR-15559
> URL: https://issues.apache.org/jira/browse/SOLR-15559
> Project: Solr
> Issue Type: Improvement
> Security Level: Public(Default Security Level. Issues are Public)
> Affects Versions: main (9.0), 8.9
> Reporter: Gus Heck
> Assignee: Gus Heck
> Priority: Minor
> Attachments: MyBenchmark.java
>
>
> While reading code for other things I noticed a small inefficiency in the
> splitSmart(String, String, boolean) method where it's creating new
> StringBuilder objects unnecessarily inside a loop. For fun I fixed this and
> JMH bench marked the fix to see what it was worth. The answer is about 30%
> relative improvement, but only a couple microseconds absolute. Not exactly
> exciting since I don't think there's anywhere that really loops hard on this,
> but having it coded and tests passing and improvements measured, might as
> well put it in.
>
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
