0

I created a simple rest service POST method that consumes a XML. I created a REST client with Jersey and created my object and i am trying to see the variations in response time by increasing my XML length, by giving larger inputs to my objects. Say, my objects takes up a simple employee details, i will increase . I see that the response time inconsistently varies , from my observation it is not dependent on the size of the xml. I am computing the time taken as follows.

long startTime = System.currentTimeMillis();

// enter code here for post `

long elapsedTime = System.currentTimeMillis() - startTime;

Please Suggest if there is a better way of doing it.

Here what i would like to get clarified is my server is in the local host and why the response time varies (say once it is 88ms and the other time it is 504ms ). What i expect is it should increase when i am giving larger inputs to my xml object but that does not happen as i observe. Please clarify or point me to a better site or book where i can read about the same.

Sara
  • 33
  • 1
  • 10

2 Answers2

0

use System.nanoTime instead of currentTimeMillis; also be aware that Java optimises on the fly and if you do not warm the server first then you will be timing interpreted execution of the byte code and then at some point it will flip to native byte code. The timings will differ by input length (in a step like function) but you have a certain amount of noise to overcome before you will see that. A lot from GC, a lot from thread scheduling and OS issues.

I recommend the mechanical sympathy group for discussions on this sort of thing. There are many discussions on how to measure these types of systems. https://groups.google.com/forum/#!forum/mechanical-sympathy

Chris K
  • 11,622
  • 1
  • 36
  • 49
0

Note that your question is quite broad (and will likely be closed as such). My explanation is similarly broad and just meant to give you some background on why you might see the behavior that you are seeing.

It is unlikely that the way you measure time makes a big difference here, given that you are up to hundreds of milliseconds. It is more likely that the web site that you are invoking is sometimes takes longer to respond.

It may help to compare it to what you see when you type a search query into Google. Sometimes the response pops up "instantaneously", but sometimes it takes a few moments to load. Since you're using the same browser for every search, it can't be the browser causing the difference.

Instead it is likely something in the web service that is varying between the calls. In the Google search example, you might be routed to a different Google server, the server might be using a different storage to search, the query might be cached at some point, etc.

So there is no way to determine why the performance is different between invocations by simply looking at the client code. Instead, you should look at the server code and profile what happens there for different invocations.

Frank van Puffelen
  • 565,676
  • 79
  • 828
  • 807