"Long running" handlers

[infectedrhytms]

I have a handler that does fairly heavy workload takes about 100ms to
execute. And the overal request per second drops drastically.

I have JMeter default setup none of my modifications.

50 users with a simple netty business logic handler and I get 4500
request/sec

I then started adding some logic to my business handler that takes about
100ms to execute and my requests/sec dropped to 19 requests/sec

Theoretically 50 users on the client machine is nothing so I bumped it to
100 users and still got 19 requests/sec

I tried a similar test using Jetty and a servlet and I can say that straight
up Netty and Jetty(servlet HTTP POST) the performance is on par. About 4500
request/sec But here is where the difference lies. In Jetty if I put that
same long running business logic, Jetty's performance drops but still dishes
out couple 1000s requests a second. But netty drops to 19 a sec. I figure
Jetty assigns a separate threat per POST request to the servlet from the
threa pool?

Is it because my netty business handler is shared somehow? Is it the the
thread pool?












-- 
View this message in context: http://n2.nabble.com/Long-running-handlers-tp4677280p4677280.html
Sent from the Netty User Group mailing list archive at Nabble.com.