Discrepancy in the transaction controller response times when number of jmeter threads are high

[Sheetal NJ]

Hi,

I am using the latest version of jmeter v 2.8 to conduct a load test for a duration of 2 hours which includes
filling in a form with details and submiting it.
In the test plan,I have grouped the transactions using a transaction controller to get the transaction response time.

Ideally there shouldnt be much difference between
the total of the response times of the samplers in a transaction and
the response time that the transaction controller gives.
The minor difference due to the clock frequnecy is fine.

However,the transaction controller's response time is inflated quite a lot compared the total of the samplers' response times.

I am mentioning the two cases that i have tested and faced this issue.

Case 1:
A load of 1000 users from 4 load generators (250 users from each).
The transaction controller's response time is inflated 10 times compared the total of the samplers' response times.

Case 2:
A load of 700 users from 2 load generators (350 users from each)
The results are favourable when i reduce the load.
The transaction response times are still inflated,but this time around its 2 times the total of the sampler's response times.

In both the cases,the IO,CPU,Memory of the servers,the load generators and the controller for load generators is monitored and there arent any issues observed.

Has anybody else faced this? I have seen few very old posts related to this but with no useful answers.
Any help would be greatly appreciated.

Thanks,
Sheetal NJ