Yes, I forgot to mention that but I've been running with
-Djavax.net.debug=ssl:handshake:verbose and sometimes -Djavax.net.debug=all.

I think I have a strong theory of what the culprit might be. I found this
post:
http://qpid.2158936.n2.nabble.com/One-SSL-SASL-EXTERNAL-queue-creation-works-with-qpidd-1-36-0-but-not-1-35-0-td7657344.html.
In the post above the user mentions the SSL_STREAM_MAX_WAIT_ms definition.
The critical log statements I mentioned in the first post also come from
SslSocket.cpp. In SslMuxSocket::accept() it calls isSslStream(fd) which
effectively determines whether the stream is SSL or Plaintext. My theory is
that in versions of Java prior to 1.8u272 the SASL/SSL/TLS functionality
must simply be faster, or results in less network traffic or something that
results in a successful negotiation with the secure broker much more
consistently than the newer versions. As I mentioned in my first post, there
were significant updates made to this functionality in 272+. It is possible
that this layer is more complex/robust but also resulted in a performance
decrease from the older versions which results in inconsistent connections
in our computing environment. 

Passing "-Djdk.tls.client.cipherSuites=<single ciper suite>" seems to at
least marginally increase performance to the point where I see more
consistent connections with the broker. I've done other things like set the
initial heap size (i.e. -Xms) to the same as the max heap size (in order to
prevent subsequent allocations) which I read online could help start up
speed.

Is there a good reason the SSL_STREAM_MAX_WAIT_ms and SSL_STREAM_MAX_RETRIES
are not configurable? I fully admit to not knowing much at a deep level
about these protocols and interactions--I'm just an application developer
who happened upon this issue in the course of his work. I'm also a bit
hamstrung testing some of these theories as my company utilizes heavily
restricted development environments.





--
Sent from: http://qpid.2158936.n2.nabble.com/Apache-Qpid-users-f2158936.html

---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to