[
https://issues.apache.org/jira/browse/CASSANDRA-13982?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16227140#comment-16227140
]
Jason Brown commented on CASSANDRA-13982:
-----------------------------------------
I put together a small, naive microbench to compare the autoboxing costs of
these functional interfaces. I ran it several times, and this result is the
best case:
{noformat}
[java] Benchmark
Mode Cnt Score Error Units
[java] AutoBoxingBench.booleanFromBooleanSupplier
sample 12695312 55.727 ± 0.581 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.00
sample ≈ 0 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.50
sample 51.000 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.90
sample 71.000 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.95
sample 76.000 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.99
sample 87.000 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.999
sample 163.000 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p0.9999
sample 10311.499 ns/op
[java]
AutoBoxingBench.booleanFromBooleanSupplier:booleanFromBooleanSupplier·p1.00
sample 1304576.000 ns/op
[java] AutoBoxingBench.booleanFromPlainSupplier
sample 12239054 60.539 ± 1.076 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.00
sample ≈ 0 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.50
sample 54.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.90
sample 75.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.95
sample 78.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.99
sample 90.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.999
sample 176.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p0.9999
sample 11632.000 ns/op
[java]
AutoBoxingBench.booleanFromPlainSupplier:booleanFromPlainSupplier·p1.00
sample 2646016.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier
sample 12538499 58.154 ± 1.302 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.00
sample ≈ 0 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.50
sample 50.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.90
sample 73.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.95
sample 76.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.99
sample 89.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.999
sample 170.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p0.9999
sample 10864.000 ns/op
[java] AutoBoxingBench.intFromIntSupplier:intFromIntSupplier·p1.00
sample 3813376.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier
sample 12102940 60.167 ± 1.524 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.00
sample ≈ 0 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.50
sample 55.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.90
sample 74.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.95
sample 79.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.99
sample 99.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.999
sample 206.000 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p0.9999
sample 11995.294 ns/op
[java] AutoBoxingBench.intFromPlainSupplier:intFromPlainSupplier·p1.00
sample 4743168.000 ns/op
{noformat}
Admittedly, most runs came out equal or the changes in this patch slightly
ahead (1-3%). I wasn't expecting much with benchmark, just to see how garbage
creation would affect the results; thus, i dropped the memory use of the bench
from 512MB to 256MB (I tried 128MB, but it was mostly the same as 256). Of
course, those differences won't make much difference in the real world or
running inside a database, but I think the incremental improvement of a little
less garbage is probably worth the effort.
Thus far, the utests look fine, and the dtests will take awhile.
> Refactoring to specialised functional interfaces
> ------------------------------------------------
>
> Key: CASSANDRA-13982
> URL: https://issues.apache.org/jira/browse/CASSANDRA-13982
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Ameya Ketkar
> Assignee: Ameya Ketkar
> Priority: Minor
> Labels: static-analysis
> Fix For: 4.x
>
> Original Estimate: 24h
> Remaining Estimate: 24h
>
> Usage of specialised functional interfaces provided by JDK, will reduce the
> autoboxing overhead hence.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
