[
https://issues.apache.org/jira/browse/CASSANDRA-6689?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13920357#comment-13920357
]
Benedict commented on CASSANDRA-6689:
-------------------------------------
bq. So the only question that I still have - what is the main goal of the
ticket? Is it to move to zero copy for memtables or minimize number of
allocations on java heap?
Zero copy would certainly be nice to retain. Otherwise why not just ref-count
the memory?
bq. allocator interface is going to be better encapsulated (basically expose
alloc/release)
I don't follow how this would improve encapsulation, sorry. Could you sketch an
example? It sounds to me like this actually would create more complexity
outside of the main body of code. It might reduce complexity a little where the
fun stuff happens, but is it all really so difficult to understand that we want
to start opting for less good solutions that require more work? Also makes it
difficult to impose tight caps on memory used, as we can't practicably
impose/manage temporary memory allocated by readers. We already have this
problem for disk reads, but no point opening up further doors for OOM.
bq. I'm not sure if we need sub pools or groups of allocators in there, I
really like how jemalloc and others do it also Netty already has implementation
of jemalloc (ByteBufAlloc) so maybe we can reuse that?...
I'm unclear how this addresses the concerns of resource management. The
ByteBufAllocator appears to simply allocate byte buffers, which by itself is a
pretty simple problem. jemalloc is also just a low level replacement to malloc,
so I'm not sure where it enters the discussion?
> Partially Off Heap Memtables
> ----------------------------
>
> Key: CASSANDRA-6689
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6689
> Project: Cassandra
> Issue Type: New Feature
> Components: Core
> Reporter: Benedict
> Assignee: Benedict
> Fix For: 2.1 beta2
>
> Attachments: CASSANDRA-6689-small-changes.patch
>
>
> Move the contents of ByteBuffers off-heap for records written to a memtable.
> (See comments for details)
--
This message was sent by Atlassian JIRA
(v6.2#6252)
