09-03-2011 01:46 AM - edited 09-03-2011 02:07 AM
I am running a test with 25 users and 18 business flows (script). My application (Java based, weblogic server) stops responding after 1 hour of test.
Error : Action.c(161): Error -27796: Failed to connect to server "172.21.164.44:6101":  Connection refused
Please note :
1) All logs are disabled.
2) Think-time, pacing are all perfect
3) I have changed setting in Runtime as : WinInet replay instead of Sockets (Windows only), still the problem persists.
4) Even the file descriptors are increased (weblogic tuning - to increase open files limit)
5) Open connections max limit is also fine.
Quick help will be highly appreciared.
09-03-2011 02:52 AM
Have you considered that your applicaiton is not responding because it actually has a problem?
You can run through some pretty simple steps to assertain this or not.
1. Does your application give errors manually at the same time you are seeing errors in the test?
2. Is your test applying the correct load? (Think requirements.)
If 1 & 2 are true then...
09-05-2011 09:34 PM - edited 09-05-2011 09:36 PM
I re-tested again yesterday with little high think time and ensured that among 25 users, there should be only 8-9 active users in the system (rest all I exited gradually).
I assumed in production, among 25 users, at any given point of time only 8-9 users will be actually doing work. Rest all might be in discussion, meetings or working on documentation.
Still, the application went down after 4 hours. Yes, I was unable to access it manually also i.e, server stopped responding.
CPU utlilization at DB server side was at times ~60% during the test. But overall, it maintained a good percentage of 90%.
I assume, VTS is same as JConsole. I did check the memory and threads there. Seems fine. Cant figure out, wats wrong!
09-06-2011 10:47 AM
90% CPU? For 4 hours?
Sounds like it did pretty well if you ask me.
Let's go back a bit here, do you have any of the following?
1. Objectives to your testing?
2. Target / Load Profile / SLAs etc.?
4. Think Time?
PS. 'Users' is not a description of load. It is not an objective or even a target. It is not a scientific meansurement of traffic. It is not useful. It is not how you should measure or describe your tests. Try getting into the habit of using phrases like 'business transactions per second', 'page views per hour', 'hits', 'end to end journeys' etc. You need this language to even begin to think about doing a good job of reporting a performance issue.
09-07-2011 01:04 PM
You mention checking memory. Can I assume you mean you checked the server console and there was still free memory available? What about heap utilization? Sounds to me like you have a memory leak in the application, and with such a low volume of users, it may be in the user session data. How large is each?
If all of these are ok, what about database connection pools? Are you freeing up database connections as they are closed?
Best of luck,
09-08-2011 02:18 AM - edited 09-08-2011 02:20 AM
Issue is resolved :)
It was because of OS issue. When we changed our UNIX box where the application code was deployed, test executed successfully for hours...
Our performance objective for the online application was: To test online application for 12 hours at a stretch. There will be a few batches that would be executed along with application.
Thanks everyone for support.