[
https://issues.apache.org/jira/browse/APEXMALHAR-2126?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15351786#comment-15351786
]
Vlad Rozov commented on APEXMALHAR-2126:
----------------------------------------
The intention of com.datatorrent.netlet.util.Slice was to avoid memory copy
when raw data is received into a buffer and then parsed into messages, not
memory allocation/deallocation. Memory for data structures/tuples still needs
to be allocated and allocating large buffer may lead to excessive memory usage
when a large buffer is allocated, but not fully used.
> Suggest: Share Slice Buffer
> ---------------------------
>
> Key: APEXMALHAR-2126
> URL: https://issues.apache.org/jira/browse/APEXMALHAR-2126
> Project: Apache Apex Malhar
> Issue Type: Improvement
> Reporter: bright chen
>
> I think the intention of Slice(com.datatorrent.netlet.util.Slice) was to
> share the buffer and avoid unnecessary memory allocation/deallocation. But
> the intension is not self-explain and lack of method to share the memory. And
> the util class org.apache.apex.malhar.lib.utils.serde.SliceUtils also create
> new memory and copy the data.
> I suggest to implement another class(Say BufferSlice), which
> - initialize buffer with relative large buffer
> - support append(byte[] data, int offset, int length)
> - dynamic reallocated buffer or throw exception when buffer is full ( based
> on the management strategy)
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
