Increase in Average Transaction Response Time and Decrease in Throughput.

[dlna]

Hi,

I ran a performance test using loadrunner for a period of 4 hours. I
noticed an increase in the average ATRT and decrease in server
throughput. Usually the increase in throughput is directly
proportional to increase the throughput. Also a decrease in available
memory was observed, the average CPU utilization(4.97%) remained low
right throughout the test period. What seems to be the issue here?


Thanks,

--
You received this message because you are subscribed to the Google "LoadRunner" group.
To post to this group, send email to LR-Loa...@googlegroups.com
To unsubscribe from this group, send email to
LR-LoadRunne...@googlegroups.com
For more options, visit this group at
http://groups.google.com/group/LR-LoadRunner?hl=en

[Shlomi Nissim]

Did you notice any out of memory errors? Any Vuser failed?

--
Shlomi Nissim
Functional Architect
HP LoadRunner and Performance Center

[chaitanya bhatt]

Your rational seems to be wrong. Response time and throughput are inversely proportional to each other.

 

-Chaitanya M Bhatt

http://www.performancecompetence.com

[Dilina Perera]

@Shlomi: These error messages were displayed.
Action.c(4): Error -27496: Internal Error - mptVUserVars (04744624) does not match ptVUserVarsFromHandledItem (054C4664) for LrwNetHandledItem at 042EF7D4

Action.c(4): Error: C interpreter run time error: Action.c (4):  Error -- memory violation : Exception ACCESS_VIOLATION received.

Action.c(131): Condition(s) of severity "ERROR" raised for a virtual user in the current PROCESS, but not necessarily in the current THREAD

Action.c(236): Warning -26627: HTTP Status-Code=404 (/common/Calendar.js) for "https://xxx.xxx.xxx.xxx:xxxx/common/Calendar.js?relversion=201005101545"

@chaitanya: An increase in throughput means that more data is sent from the server to the client doesn't this increase the average transaction response time?

--
Dilina Perera

[vikrant dixit]

It seems that no of Vuser goes down .Check the Running vuser graph corresponding to response time.

[Krishnakanth PPS]

Dilina.

When your throughput during a test decreases, your response time increases, and when your throughput increases your response time would be relatively good.

Since more data is being sent through the server, thus the server is able to process requests faster, and hence your response time would be relatively lower.

If throughput is low, then there is some bottleneck which is causing the response time to increase. 

So, response time and throughput are inversely proportional. Check out the below links for a better understanding.

http://docs.sun.com/source/819-4741/images/res-time.gif 

http://docs.sun.com/app/docs/doc/819-4741/6n6sei5ra?a=view

As for the errors, I havent seen such errors before, but there seems to be some conflict or a deadlock for a resource. 

Make sure you run in VUGen first without any errors and then run a smoke test with 5-10 users for a period of 5-10 mins and see if you see these errors. 

Do let us know your results. 

Thanks.

Krishna Kanth

On Tue, May 11, 2010 at 12:48 PM, Dilina Perera <dln...@gmail.com> wrote:

[kranthi reddy]

Hi,

First know about to knw that, there is no correlation between system response time and the server utilization. The response time increase might happen even when the server is less utilized. Increase in RT is caused might because of queueing of the requests. The user arrival pattern contributes to the high response time.

and waiting time + Processing time = Response Time.

So if waiting time is more then obviously throughput will decrease as server cant process the requests.

U need to check the memory leaks and no.of sessions that the application can handle.

Regards,

Kranthi

[chaitanya bhatt]

From:On Tue, May 11, 2010 at 1:18 AM, Dilina Perera <dln...@gmail.com> wrote:

>@chaitanya: An increase in throughput means that more data is sent from the server to the >client doesn't this increase the average transaction response time?

 

No. Think again.

 

And regarding those memory access violation errors - check your protocol against the process/thread option available in the run time settings. If you are using a certain protocol which expects you to run the script as a process, but instead you are running it as a thread then you might see such errors. Also, keep an eye on the host machine system resources. Because, a choked host can cause vusers to throw such errors.

 

And those 404 errors are "authetication failed" errors. Double check and see if you have correlated the right dynamic strings.

-Chaitanya M Bhatt

http://www.performancecompetence.com

[Dilina Perera]

@ kranthi: All the Vuser(90) are initialized at the start and they all go down at the end of the test. Therefore the user arrival pattern doesn't affect and it clearly isn't the reason for the increase RT.

[Floris Kraak]

On Tue, May 11, 2010 at 10:25 AM, Dilina Perera <dln...@gmail.com> wrote:
> @ kranthi: All the Vuser(90) are initialized at the start and they all go
> down at the end of the test. Therefore the user arrival pattern doesn't
> affect and it clearly isn't the reason for the increase RT.
>

You may be overloading that application from the start though.

Have you tested this with a very slow ramup? Add virtual users slowly,
preferably in stages. 1 virtual user per minute, ramp up to 10, run
for 30 minutes, ramp up further to 20, etc. If there is a (memory)
leak of some kind your available memory pool should decrease with 10
virtual users as well. If on the other hand it's a case of some kind
of queuing or thread/connection pools filling up response times may
stay stable at 10 virtual users.

I agree chaitanya bhatt btw - some of those errors are not application
failures, they're errors in your script and/or setup. Please check if
you get those errors with a single (yes, 1) virtual user as well. If
so: Fix that first before you proceed.

Regards,
Floris
---
'What does it mean to say that one is 48% slower? That's like saying
that a squirrel is 48% juicier than an orange - maybe it's true, but
anybody who puts the two in a blender to compare them is kind of
sick.'
--- Linus Torvalds

[kashif awan]

I agree to this thought. They are inversely proportional. If the TRT (Trx. Resp. Time) is increasing, so it means that server is responding late to a particular request so throughput is always low while TRT is increasing.

 

Regards,