The.Algorithm.Design.Manual.Springer-Verlag.1998

Dictionaries
Bottom line: Which binary search tree is best for your application? Probably the balanced tree for
which you have the best implementation readily available. See the choices below. Which flavor of
balanced tree is probably not as important as how good the programmer was who coded it.
● B-trees - For data sets so large that they will not fit in main memory (say more than 1,000,000
items) your best bet will be some flavor of a B-tree. As soon as the data structure gets outside of
main memory, the search time to access a particular location on a disk or CD-ROM can kill you,
since this is several orders of magnitude slower than accessing RAM.
The idea behind a B-tree is to collapse several levels of a binary search tree into a single large
node, so that we can make the equivalent of several search steps before another disk access is
needed. We can thereafter reference enormous numbers of keys using only a few disk accesses. To
get the full benefit from using a B-tree, it is important to understand explicitly how the secondary
storage device and virtual memory interact, through constants such as page size and virtual/real
address space.
Even for modest-sized data sets, unexpectedly poor performance of a data structure may be due to
excessive swapping, so listen to your disk to help decide whether you should be using a B-tree.
● Skip lists - These are somewhat of a cult data structure. Their primary benefits seem to be ease of
implementation relative to balanced trees. If you are using a canned tree implementation, and thus
not coding it yourself, this benefit is eliminated. I wouldn't bother with them.
Implementations: LEDA (see Section ) provides an extremely complete collection of dictionary data
structures in C++, including hashing, perfect hashing, B-trees, red-black trees, random search trees, and
skip lists. Given all of these choices, their default dictionary implementation is a randomized search tree
[AS89], presumably reflecting which structure they expect to be most efficient in practice.
XTango (see Section ) is an algorithm animation system for UNIX and X-windows that includes
animations of such dictionary data structures as AVL trees, binary search trees, hashing, red-black trees,
and treaps (randomized search trees). Many of these are interesting and quite informative to watch.
Further, the C source code for each animation is included.
The 1996 DIMACS implementation challenge focused on elementary data structures like dictionaries.
The world's best available implementations were likely to be identified during the course of the
challenge, and they are accessible from http://dimacs.rutgers.edu/ .
Bare bones implementations in C and Pascal of a dizzying variety of dictionary data structures appear in
[GBY91], among them several variations on hashing and binary search trees, and optimal binary search
tree construction. See Section for details.
file:///E|/BOOK/BOOK3/NODE129.HTM (4 of 5) [19/1/2003 1:30:03]