[ 
https://issues.apache.org/jira/browse/CASSANDRA-4937?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13577878#comment-13577878
 ] 

Pavel Yaskevich commented on CASSANDRA-4937:
--------------------------------------------

bq. preheats row data if 90% of the rows in the sstable are under the page size 
that we're fadvising.

I see the reason to do that if we have big rows (index promoted to the Index 
component so we don't touch first page of a row) and we don't know where we 
would be hitting them but this is why I don't think that 90% is a good idea 

  - we don't know distribution of those big rows so if we small row which was 
sharing page with big row it's still good to preheat as we read on page basis.

  - if we still preheat first page that we didn't need it would actually be 
migrated by kernel automatically with adoptive read-ahead for example.

  - if rows grow over time it would be a sadden change (flip-flop) in 
behavior/latencies.

  - even if 90% are bigger of the page size it's quiet possible that keys that 
we actually migrated in the cache are in other 10%.
                
> CRAR improvements (object cache + CompressionMetadata chunk offset storage 
> moved off-heap).
> -------------------------------------------------------------------------------------------
>
>                 Key: CASSANDRA-4937
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-4937
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: Pavel Yaskevich
>            Assignee: Pavel Yaskevich
>              Labels: core
>             Fix For: 1.2.2
>
>         Attachments: 4937-v3.txt, CASSANDRA-4937.patch, 
> CASSANDRA-4937-trunk.patch
>
>
> After good amount of testing on one of the clusters it was found that in 
> order to improve read latency we need to minimize allocation rate that 
> compression involves, that minimizes GC (as well as heap usage) and 
> substantially decreases latency on read heavy workloads. 
> I have also discovered that RAR skip cache harms performance in situation 
> when reads are done in parallel with compaction working with relatively big 
> SSTable files (few GB and more). The attached patch removes possibility to 
> skip cache from compressed files (I can also add changes to RAR to remove 
> skip cache functionality as a separate patch). 

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Reply via email to