[ 
https://issues.apache.org/jira/browse/HADOOP-15033?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16251631#comment-16251631
 ] 

Dmitry Chuyko edited comment on HADOOP-15033 at 11/17/17 9:24 AM:
------------------------------------------------------------------

It worths mentioning here that java.util.zip pure Java implementation compiled 
by C2 without some inlining is ~2x slower for len=512 than Hadoop's pure Java 
though they basically use the same approach to calculation. It is a subject for 
improvement in core library for platforms and JVMS that don't provide CRC 
intrinsics. But again in real life on Hotspot on x86 and aarch64 you always get 
an intrinsic implementation.


was (Author: dchuyko):
It worths mentioning here that java.util.zip pure Java implementation compiled 
by C2 is ~2x slower for len=512 than Hadoop's pure Java though they basically 
use the same approach to calculation. It is a subject for improvement in core 
library for platforms and JVMS that don't provide CRC intrinsics. But again in 
real life on Hotspot on x86 and aarch64 you always get an intrinsic 
implementation.

> Use java.util.zip.CRC32C for Java 9 and above
> ---------------------------------------------
>
>                 Key: HADOOP-15033
>                 URL: https://issues.apache.org/jira/browse/HADOOP-15033
>             Project: Hadoop Common
>          Issue Type: Improvement
>          Components: performance, util
>    Affects Versions: 3.0.0
>            Reporter: Dmitry Chuyko
>
> java.util.zip.CRC32C implementation is available since Java 9.
> https://docs.oracle.com/javase/9/docs/api/java/util/zip/CRC32C.html
> Platform specific assembler intrinsics make it more effective than any pure 
> Java implementation.
> Hadoop is compiled against Java 8 but class constructor may be accessible 
> with method handle on 9 to instances implementing Checksum in runtime.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to