[ 
https://issues.apache.org/jira/browse/CASSANDRA-9738?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14723917#comment-14723917
 ] 

Ariel Weisberg commented on CASSANDRA-9738:
-------------------------------------------

I agree there is overlap in. The scope of what we need to make the key cache 
work is smaller, but it might be game over for the key cache and large 
partitions anyways even with Robert's improvements. Constantly loading and 
unloading large partitions is never going to work. We are at least reaching 
parity with the old key cache.

If we could map the entire partition index then you could point this code at 
the mapping and it would work. It would not be ideal because what we are doing 
is a basic translation of the object graph which is an array pointing to 
variable size objects. Binary searches are going to result in access to two 
pages for each comparison. Possibly still better than what we have today?

I think the trick right now is figuring out what we can do for 3.0, and what 
the next intermediate step is. If Robert goes and makes this work off-heap then 
the key cache can maybe go off heap for 3.0. Seems like removing the key cache 
and reliably operating against a memory map is unlikely for 3.0, but maybe 
shortly after.

What would the timeline be for 9754? That kind of determines what intermediate 
steps make sense.

> Migrate key-cache to be fully off-heap
> --------------------------------------
>
>                 Key: CASSANDRA-9738
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9738
>             Project: Cassandra
>          Issue Type: Sub-task
>            Reporter: Robert Stupp
>            Assignee: Robert Stupp
>             Fix For: 3.0 beta 2
>
>
> Key cache still uses a concurrent map on-heap. This could go to off-heap and 
> feels doable now after CASSANDRA-8099.
> Evaluation should be done in advance based on a POC to prove that pure 
> off-heap counter cache buys a performance and/or gc-pressure improvement.
> In theory, elimination of on-heap management of the map should buy us some 
> benefit.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to