[
https://issues.apache.org/jira/browse/QPID-3028?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Robbie Gemmell updated QPID-3028:
---------------------------------
Status: Ready To Review (was: In Progress)
> Java broker uses excessive amounts of memory to service 0-10 connections
> ------------------------------------------------------------------------
>
> Key: QPID-3028
> URL: https://issues.apache.org/jira/browse/QPID-3028
> Project: Qpid
> Issue Type: Bug
> Components: Java Broker
> Affects Versions: 0.6, 0.8
> Reporter: Robbie Gemmell
> Assignee: Robbie Gemmell
> Priority: Critical
> Fix For: 0.9
>
>
> The Java broker uses excessive amounts of memory to service 0-10 connections.
> There are a number of reasons for this, such as:
> - Use of an arbitrary 64000 element array in the Assembler to store
> incomplete incoming per-channel data.
> - Use of per-connection ThreadLocals for the BBDecoder in the Assembler
> interacting with the thread pool threads moving between connections, leading
> to an NxM number of BBDecoders in use.
> - Use of per-connection ThreadLocals for the BBEncoder in the Disassembler
> interacting with the thread pool threads moving between connections, leading
> to an NxM number of BBEncoders in use.
> - BBDecoder maintaining a hard reference to the last ByteBuffer it used.
> - The binary->String cache within BBDecoder maintaining a hard reference to
> the entire backing array for the network read buffer which contained the
> cached binary, rather than just the bit being cached.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
---------------------------------------------------------------------
Apache Qpid - AMQP Messaging Implementation
Project: http://qpid.apache.org
Use/Interact: mailto:[email protected]
