cassandra-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Branimir Lambov (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CASSANDRA-11541) correct the java documentation for SlabAllocator and NativeAllocator
Date Wed, 04 May 2016 09:31:12 GMT

    [ https://issues.apache.org/jira/browse/CASSANDRA-11541?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15270378#comment-15270378
] 

Branimir Lambov commented on CASSANDRA-11541:
---------------------------------------------

It is not very clear which size is less than 128k in both comments, I would change that phrase
to "allocations of up to 128kb". The 1MB is fixed, "by default" should be removed, perhaps
something like:
{quote}
...
large (1MB) global regions and then doles them out to threads that request
smaller sized (up to 128kb) slices into the array.
{quote}

I believe you mean "exponentially" instead of logarithmically (since for step/scale _n_, size
is 2^_n_) in the second comment, I would change it to:
{quote}
This NativeAllocator uses global slab allocation strategy with slab size that
scales exponentially from 8kb to 1Mb to serve allocations of up to 128kb.
<p></p>
The slab allocation reduces heap fragmentation from small long-lived objects.
{quote}

> correct the java documentation for SlabAllocator and NativeAllocator
> --------------------------------------------------------------------
>
>                 Key: CASSANDRA-11541
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-11541
>             Project: Cassandra
>          Issue Type: Improvement
>            Reporter: ZhaoYang
>            Assignee: ZhaoYang
>            Priority: Trivial
>              Labels: document, lhf
>             Fix For: 2.2.x
>
>         Attachments: CASSANDRA-11541.patch
>
>
> I heard a lot that Cassandra uses 2MB slab allocation strategy for memtable to improve
its memory efficiency. But in fact, it has been changed from 2MB to 1 MB. 
> And in NativeAllocator, it's logarithmically from 8kb to 1mb.
> large number of tables may not cause too much trouble in term of memtable.



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

Mime
View raw message