问题
I am writing a simple codec. The tree will be precomputed and there wont be any changes once its build. It will just be searched.
All the leaf nodes of a balanced binary tree are the signal values and internal nodes are the approximated compressed representations.
Is a list implementation using stl vector scalable if I have a large value of leaf nodes? Currently I dont know how large large is.
List implementation e.g. 1,2,3,4,5,6,7 if I have 4 leaf nodes
then children of
root(1)-> 2,3
2->4,5
3->6,7
so I can simply jump to the children using their position in the vector.
回答1:
Using a "list" or "array" should not pose any issue in this case (since the tree is immutable). The only requirement for O(log n) search is quick random access (ie, O(1) access to a given index).
You can use either a vector or a deque, both are suitable. You might run in a scalability issue with the vector if the system cannot find a large enough memory block to hold all elements, though to begin with it should prove simpler. If you hit this roadblock, switch to deque, and although you might lose some speed, it should allow you to grow further due to its fragmented nature.
回答2:
I would prefer a preallocated array of your tree nodes (mostly malloced at the initialization time ) with MAX_ELEMENTS as opposed to a stl vector for simple reason of having all tree nodes continuously located on the heap for quicker access at run time.
I said quicker access mainly because the pointer jumps may not be uniform with a distributed stl vector elements for a really huge number of elements in the list (say overflowing a 32 bit number )
You can look at it from the cache line misses from the L1 caches access Point of view of you processor.
来源：https://stackoverflow.com/questions/10715389/is-a-list-implementation-of-binary-tree-scalable