| || |
Client Server testing
My manager and I disagreed about the answer for this question, therefore i told him the best way to get the right answer we will post the question here:
What is the different between client server testing and other type of testing? what is the most important aspect in performance testing in client server? is it the network performance? if the newwork performance is the answer what is the most aspect of the network performance?
Re: Client Server testing
For differences in types of testing: http://www.qaforums.com/Forum15/HTML/002718.html
As far the most important... almost any time you say most important you must ask "important to whom?" Importance is ranked by corporate priorities, which are different for most corporations. http://www.context-driven-testing.com/
Re: Client Server testing
Usually the answer to your question has poilitical overtones. Clearly to the users the most important aspect is application performance. The user really could care less whether there is an antennae (sp?) or a pice of glass fiber connecting them to the corporate network, so long as the application responds in a timely fashion. Note, usually performance standards are built upon the basis of application performance.
However, when performance suffers, invariably people say "the network is slow" when it may just be the application. Not knowing where in the OSI stack the problem lays (above or below the session layer) has lead to lots of saber rattling and finger pointing between application management and network management groups in the past. You may have even been in a room as a witness to some of these discussions.
Here's how I try to resolve the problem. First I test the application in a lab environment with the fastest, cleanest network connections possible. I want to make sure the application scales uncolored by the network.
At the same time as the tuning in the lab test for the application is goign on, I use a session layer network load testing tool on the corporate network. I derive a model for the network load test tool conversations from the pattern of sends and receives generated by the business processes under examination in the lab. Incidentally, if you use LoadRunner, record a business process in WinSock and look at the recording log - you should see a note on the pattern of sends and receives I would use. I use and recommend a tool called Chariot from NetIQ for these network load tests. It is not the only tool on the market, but it does work the best for me. As a full-disclosure note, I do not sell Chariot, but I do deliver services for the tool. These network load tests will allow the networking group to test and tune the network prior to the first bit of new application traffic flows across the wire. Any problems should be resolved prior to the application leaves the lab.
The next step is dependent upon what you have in the lab. If you have the Cloud or Storm products from Shunra (once again, I do not sell or deliver services for these products) then use them to model your network and re-run your application load tests with the delays appropriate to your network in place. This should provide a very accurate measure of expected performance. If you do not have the Cloud or Storm, but do have a traffice generator, such as Chariot, then model your highest speed and lowest speed links. Re-run your tests with an appropriate number of users across each link. Ramp up the users to a full load. Next, add additional traffice across your lowest speed and highest speed links until they reach their historical worst utilization levels. This should provide you a perspective on the maximum out of tollerance condition that users would expect. Ramp the network traffice generator load back down to the average current utlitization and you should find what is the expected average response time for the application on the network.
If you're the network admin and you're going to be blamed for poor performance all of the time, then network performance is the critical measurement. If you're the application manager and your bonus is tied to the performance of the application, you are going to be concerned with application performance, even if that means the network group has to absorbe the additional cost of laying fiber to every desktop, Darnit!
I hope this has been helpful,
iTest Solutions, Inc
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