[
https://issues.apache.org/jira/browse/QPID-2410?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15167294#comment-15167294
]
Cliff Jansen commented on QPID-2410:
------------------------------------
The reserved byte space at the front and end of the QPID buffer is stripped at
the other end, so it does not alter the frame size seen by the application code
that processes the AMQP frame.
The fact of the mater is that the natural and efficient handling of application
data streams is very different in NSS versus SChannel (transform via copy
versus transform in place).
In Proton, small frames are handled as they are in QPID (in place) and for
larger frames, only the front portion is transformed in place with the rest
getting copied (its actually uglier than that but that's the general idea).
For correctness in QPID with large frames on Windows, something along those
lines is probably the simplest thing to be done here, where "simple" is a
relative term (buffer breakup, perhaps single completions mapped to multiple
ones).
For best performance with large frames, some deeper thinking is warranted. I
seem to recall that some sort of scatter gather fanciness lurks somewhere in
SChannel, but as always with Windows, finding the set of APIs that work across
the desired versions of Windows is likely a challenge.
> perftest hang in SSL on Windows with large buffers
> --------------------------------------------------
>
> Key: QPID-2410
> URL: https://issues.apache.org/jira/browse/QPID-2410
> Project: Qpid
> Issue Type: Bug
> Components: C++ Client
> Affects Versions: 0.7
> Environment: Windows client, Linux broker
> Reporter: Cliff Jansen
> Assignee: Steve Huston
>
> The following command:
> perftest --count 1 --size 102400 -P ssl --port 5671 --broker linuxhost
> --username testuser --password secret --mechanism PLAIN
> hangs on a Windows client. Reducing the message body size to 1024 bytes
> makes the hang go away, as does turning off SSL.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
