[ 
https://issues.apache.org/jira/browse/DERBY-5503?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13150641#comment-13150641
 ] 

Mike Matrigali commented on DERBY-5503:
---------------------------------------

Derby  pays the encryption cost either when a page comes off disk and goes into 
the derby cache, or when a dirty page moves from cache
to disk.   There is also an encryption cost for every log record written, which 
would be a normal part of a benchmark.  There is also a cost for
every log record read, but this cost is rarely measured in a benchmark as it 
would only happen during abort transaction and/or database recovery at startup 
after an abnormal exit.  
So worst case for this comparison if you can come up with an experiment where 
the No encryption case is running 100% cpu on 
all available processors per thread, and the size of the experiment does not 
get a lot of derby cache hits.

Update benchmarks are hard to get consistent reproducible results unless you 
run for a long time as derby separates the I/O(and associated
encryption cost) from the transaction that does the update.  Then your results 
depend whether you happen to hit a checkpoint while you are
measuring or not.

A select benchmark that loads the data as part of the setup probably gives fair 
results across each run.   For worst case you should pick the
size of the data so it is likely to sit in the OS cache after creation, thus 
making the derby cache faults from disk really not be I/O's thus making
the test cpu bound.  On the other hand one could come up with a best case 
scenario where all operations happen in cache and there should
be no degredation.  

                
> Measure the performance degradation incurred by encrypting Derby databases
> --------------------------------------------------------------------------
>
>                 Key: DERBY-5503
>                 URL: https://issues.apache.org/jira/browse/DERBY-5503
>             Project: Derby
>          Issue Type: Task
>          Components: Store
>    Affects Versions: 10.9.0.0
>            Reporter: Rick Hillegas
>
> It would be good to measure the performance degradation incurred by Derby's 
> encryption.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to