[ 
https://issues.apache.org/jira/browse/IMPALA-6645?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16477561#comment-16477561
 ] 

ASF subversion and git services commented on IMPALA-6645:
---------------------------------------------------------

Commit e4fcc31cf1385fa99c06a04c0c4edecfbf5dbd95 in impala's branch 
refs/heads/2.x from [[email protected]]
[ https://git-wip-us.apache.org/repos/asf?p=impala.git;h=e4fcc31c ]

IMPALA-6645: Enable disk spill encryption by default

Perf:
Targeted benchmarks with a heavily spilling query on a machine
with PCLMULQDQ support show < 5% of CPU time spent in encryption and
decryption. PCLMULQDQ was introduced in AMD Bulldozer (c. 2011)
and Intel Westmere (c. 2010).

Testing:
Ran core tests with the change.

Updated the custom cluster test to exercise the non-default
configuration.

Change-Id: Iee4be2a95d689f66c3663d99e4df0fb3968893a9
Reviewed-on: http://gerrit.cloudera.org:8080/10345
Reviewed-by: Tim Armstrong <[email protected]>
Tested-by: Tim Armstrong <[email protected]>


> Consider enabling disk spill encryption by default
> --------------------------------------------------
>
>                 Key: IMPALA-6645
>                 URL: https://issues.apache.org/jira/browse/IMPALA-6645
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Backend
>    Affects Versions: Not Applicable
>            Reporter: Tim Armstrong
>            Assignee: Tim Armstrong
>            Priority: Major
>              Labels: resource-management
>
> We currently have integrity checks disabled by default because of the 
> additional CPU cost. 
> Now that IMPALA-6219 is in, the integrity checks are cheaper so it may be 
> worth just paying the cost. We could also consider doing a lighter-weight CRC 
> checksum, but AES-GCM may be fast enough at 2.3GB/s to do by default.
> We should do an experiment to see if there's a major impact on spill-to-disk 
> perf.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

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

Reply via email to