| || |
Does LoadRunner have a problem with 10Gbps Windows interface?
Just joined, first post. Hopefully someone can help.
My Controller/Load Generator is installed with a 10Gbps interface, and I have proven, outside of LoadRunner, that it can send traffic in both directions to my OHS webserver (OHS is Oracle's implementation of Apache). The webserver is on an Exalogic node.
But, it appears that I can only send LoadRunner load at around 1Gbps. I've looked at and monitored all the usual suspects, including the Run-time setting Speed Simulation (set to Use maximum bandwidth).
Is there any known issue or configuration I have to change to make this work?
LoadRunner doesnt care what the transport layer looks like. All of it's services are in the layers 6-7 of the OSI model, layered on top of TCP and UDP.
The controller itself would never generate 10gbs of throughput when communicating to the load generators. If you are using the controller as a load generator then you will want to rethink this solution, for it is considered a poor practice and should not be used. Always use independent load generators from the controller. I generally go for a minimum of three.
As to load generators filling a 10gbs pipe, that likelihood is low, just due to the nature of the client server model which has delays between requests coming from users. Even quite heavy commercial sites do not have this type of outward link speed to the internet, for this speed is slightly above OC-192 in speed and would most likely be used for an internet backbone as opposed to a single termination link speed.
What a load generator can generate on load is a very complex topic. You here have noted pipe capacity, but each virtual user has some amount of discrete and finite resource that it leverages not only on network, but also CPU, RAM and DISK. Exceed capacity in any of the finite bottleneck areas and there is really not any more that the generator will do. It is quite probable you have hit a limit on RAM if you are running a lot of users (keep in mind that the load generator is a 32 bit process and has a limit of 4GB of addressable RAM or DISK if you have decided to log extensively. A bottleneck on CPU rarely comes into play except when users are run without think time or pacing under the mistaken theory that ~n~ number of user would then represent some multiple of that number. Why this fails is because eliminating think time violates the client server model which expects inter-request delays from the same client, during which time the requests of other sessions are addressed.
It is also possible that you have hit the limit of your application. Unless you are monitoring the finite capacity elements of all of your application under test components and load generation/test bed components you will never know.
Recommended: Move your controller to a 1gbs or 100mbs link. Your 10gbs link is wasted on the controller.
Replace ineffective offshore contracts, LoadRunnerByTheHour
. Starting @ $19.95/hr USD.
Put us to the test, skilled expertise is less expensive than you might imagine.
Twitter: @LoadRunnerBTH @PerfBytes