Jump to content

Talk:AVL Tree/Performance

From Rosetta Code

The D Version on the main page will also have O(N2) performance. It too descends the tree during balancing.NNcNannara (talk) 23:06, 15 July 2016 (UTC)

Note that well written Managed Code is quite rapid. The Native C++ Code is just a tad faster.NNcNannara (talk) 23:09, 15 July 2016 (UTC)

From the first performance test in C++, it is easy to see why a lot of people think that AVL is slower - because of the rotations. The second test in C# of Red/Black vs AVL reveals the truth. Properly written rotations are very fast. AVL does a few more rotations but these rotations have very high performance (just exchanging a few pointers or references). The overall result is that AVL is superior. NNcNannara (talk) 10:29, 16 July 2016 (UTC)

Cookies help us deliver our services. By using our services, you agree to our use of cookies.