You are viewing a plain text version of this content. The canonical link for it is here.
Posted to proton@qpid.apache.org by "Justin Ross (JIRA)" <ji...@apache.org> on 2015/11/30 15:22:10 UTC

[jira] [Closed] (PROTON-998) [C++ binding] Restore performance with lazy message creation and reuse

     [ https://issues.apache.org/jira/browse/PROTON-998?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Justin Ross closed PROTON-998.
------------------------------
    Resolution: Fixed

> [C++ binding] Restore performance with lazy message creation and reuse
> ----------------------------------------------------------------------
>
>                 Key: PROTON-998
>                 URL: https://issues.apache.org/jira/browse/PROTON-998
>             Project: Qpid Proton
>          Issue Type: Bug
>    Affects Versions: 0.10.1
>         Environment: C++ reactive binding
>            Reporter: Cliff Jansen
>            Assignee: Cliff Jansen
>             Fix For: 0.10.1
>
>
> The C++ binding has become several times slower over time.
> Investigating this performance degradation yields the following
> observations.
> Creating and destroying some Proton C objects is expensive.
> A pn_data_t requires 7 separate malloc and free calls.
> A pn_message_t requires almost 60 (there are 7 pn_data_t members in a
> pn_message_t)
> The cost set-up and tear down can exceed the use cost, so re-use can
> provide a serious performance win.
> As a corollary, no Proton object should be included in a C++ messaging
> event for the 'just in case' scenario, since so many are ignored by
> the application.
> Sample performance numbers for loopback messages on a 4 core 8 thread laptop:
> 62,000 msgs/sec:   unused message per event, except during on_message (current code)
> 135,000 msgs/sec:  lazy message creation and destruction at on_message
> 190,000 msgs/sec:  re-use pn_message_t object for all inbound messages on a connection



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