Hi.
I am currently load testing a Comet application running on Glassfish 2.1.
When running longer tests I noticed that the CPU was steadily rising. After
a while the CPU would hit 100% and then OOM's etc. would start to happen. I
started to shave off unknown parts until I could get pinpoint what causes
this, and the result is as below.
This is my web application (Servlet), the simplest possible implementation:
public class HelloWorldServlet extends HttpServlet {
@Override
protected void doGet(HttpServletRequest req, HttpServletResponse resp)
throws ServletException, IOException {
resp.getWriter().write("Hello world");
}
}
The servlet does not take advantage or reference Comet at all. With this
application deployed we applied 200 request/second for the HelloWorld.
This is the CPU usage with Comet disabled:
http://www.nabble.com/file/p23922811/cpu_comet_disabled.png
This is the CPU usage with Comet enabled:
http://www.nabble.com/file/p23922811/cpu_comet_enabled.png
It is pretty obvious that the CPU usage is much higher with Comet enabled
and also show a growing trend. If we plot the CPU with the heap usage we get
this:
http://www.nabble.com/file/p23922811/cpu_comet_enabled-cpumem.png
It seems that the CPU follows the heap size, but does not recover completely
with the gc's.
I am going to perform some more analysis, but I would like to know if this
is known/documented/expected etc. behavior?
I can turn this on and off by enabling/disabling Comet. Am I misusing Comet
by doing this? Do I have to have special Comet handling in my servlet if I
turn it on? If no, then this seems like inproper behavior, if yes, then
what?
--
View this message in context: http://www.nabble.com/Comet---growing-CPU-usage-tp23922811p23922811.html
Sent from the Grizzly - Users mailing list archive at Nabble.com.
