[
https://issues.apache.org/jira/browse/DRILL-128?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13806846#comment-13806846
]
John Morris edited comment on DRILL-128 at 10/28/13 3:28 PM:
-------------------------------------------------------------
We'd need to clarify our shared memory model. For example,
o Can the same segment be attached at the same address in all prodesses?
(allows embedded pointers vs offsets)
o Platform specific page size (Huge or otherwise)
(necessary on x86 to keep page tables to a reasonable size)
Given that pointers don't work in a Java environment, there probably isn't need
for a global attach address.
was (Author: coyotebush):
We'd need to clarify our shared memory model. For example,
o Can the same segment be attached at the same address in all prodesses?
(allows embedded pointers vs offsets)
o Platform specific page size (Huge or otherwise)
(necessary on x86 to keep page tables to a reasonable size)
> Shared Memory Toolkit
> ---------------------
>
> Key: DRILL-128
> URL: https://issues.apache.org/jira/browse/DRILL-128
> Project: Apache Drill
> Issue Type: New Feature
> Reporter: Jacques Nadeau
>
> Because of the focus on the memory format of ValueVectors, Drill could
> benefit from being able to operate across processes using shared memory.
> This would be useful in various storage engines such HBase and could also be
> useful where Drill might be running unsafe code such as when running an
> untrusted UDF. Rather than requiring each implementer to build their own
> interfaces, we should provide some higher level shared tools for this purpose.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
