测量Java算法的性能

Question

How can i measure the performance of my java algorithm effectively ? is there any accurate way of doing it?

i read other questions of same kind but not satisfied. any help would be appreciated.

Answer 1

long reference=System.nanoTime();
your_funct();
long finishm=System.nanoTime();

System.out.println( ( (double)(finishm-reference) )/1000000000.0);  //in seconds

Has a meaningful level of ~0.003 seconds in my machine. I mean, you measure in nano-seconds but smallest step is around 3000000 nanoseconds in my machine.

Answer 2

You ask for performance which indicates some sort of timing. But then what would you compare against?

A general way of measuring an algorithm is using Big O, which takes a simplified mathematical approach.

To explain this at a very basic level a simple linear search of a list of integers has a linear (n) worst case big o. Eg: for(int i = 0; i < sizeofarray; ++i) if(array[i] == to_find) return i;

At the worst case this would take i iterations (often number is referred to as n in big o) - so we call it n or linear complexity algorithm.

Something like a bubblesort algorithm is a loop within a loop so we have n * n complexity = n^2 or quadratic complexity.

Comparing like with like, if we just consider sorting, quicksort is more efficient than quadratic complexity (it is n log n complexity) so you can consider quicksort being 'better' than bubblesort.

So when evaluating your algorithm think about it in terms of n. Is there a loop? how many? The fewer the better. No loops even better - constant big o.

Answer 3

If can make it practical or theoretical. If practical then put a timer before the algorithm starts and stop it when it ends. If theoretical then use Big O notation (isn't that hard) and you'll get an estimate of its time or space complexity.

Answer 4

You can use some profilers. Many IDEs (such as Netbeans) have one.

Answer 5

最好的方法仍然是java.lang.System.currentTimeMillis（），因为无论您使用的是哪种IDE，它都可以工作。