Robert Greig wrote:
2008/5/20 Rajith Attapattu <[EMAIL PROTECTED]>:
Actually the java JMS client on trunk is quite stable and the JMS Client/C++
broker can give the same (if not better) performance than the M2.1 java
broker.
Do you have any comparative numbers, e.g. running anything in perftests?
RG
Robert,
This is not scientific, but I see Java on one pub/sub 52% slower than
C++ client. I think
we can be closer that to ~ 40% for the 1:1 case given a test where
Arnuad just piped the data
to the socket to measure theoretical max for java.When you start
bringing up more sessions
from a single client, then the difference between the C++ and Java
client seems to spread
even more. From the last data I saw, I think the Java client 0-10 &
0-9/0-8 codepaths are
about equal (within 10% variance) in throughput currently for 1:1 pub/sub
Carl.
