We have just done a migration from the commercial SGES v9.1 (GF 2.1.1) to
Glassfish 3.1.1.
Immeditately we noticed a significant drop in performance.
We've configured the GF 3.1.1 in the same manner as the SGES (i.e. same
memory, GC, thread sizes, http requestor instances..etc etc).
And of course we set the JVM to -server as opposed to -client which is the
default setting for Glassfish.
For the sake of fault isolation I did a simple test servlet that only returns
a static string for each doGet() operation.
On the same HW with the same test application there is a ~1000tps difference
between the SGES and the Glassfish.
And that's only running the test for like a minute so any GC settings are not
yet affecting.
The CPU usage is the same so the Glassfish has to do some extra job for each
request.
Is it supposed to be like this?
What else am I missing?
Is there any other development mode setting (apart from -client) that should
be configured?
Bear in mind that the test so far has been with a simple servlet returning a
static string which would be the most stupid/simple app to execute.
--
[Message sent by forum member 'DMoniX']
View Post: http://forums.java.net/node/873250