This library provides an experimental trie implementation using Bonsai structures, described in the following articles:
- Darragh, Cleary and Witten, "Bonsai: A Compact Representation of Trees", SPE, 1993.
- Poyias and Raman, "Improved Practical Compact Dynamic Tries", SPIRE, 2015.
The former and latter are implemented by the BonsaiDCW and BonsaiPR classes, respectively. BonsaiPR provides a very simple m-Bonsai (recursive) implementation without the second Bonsai hash table to store displacement values.
I consulted the mBonsai implementation.
The experiments were carried out on Quad-core Intel Core i7 @4.0 GHz, with 16 GiB of RAM (L2 cache: 256 KiB; L3 cache: 8 MiB), running Mac OS X 10.12. The codes were compiled using Apple LLVM version 8 (clang-8) with optimization -O3. The runtimes were measured using std::chrono::duration_cast. To measure the required memory sizes, the /usr/bin/time command was used. The tries were constructed from sampled geographic names on the asciiname column from the GeoNames dump (# of nodes: 8,575,826, # of keys: 1,000,000, raw size: 14.9 MiB).
To explore BonsaiPR parameters for constructing compact trie structures, some combinations of two parameters were tested. One is the load factor of a hash table. The other is each primary slot size for storing displacement values, which is the third parameter of the constructor, width_1st.
From the results, pairs (0.8, 6) and (0.9, 8) are good parameter combinations.
Both of BonsaiDCW and BonsaiPR were implemented in 0.8 and 0.9 load factors. BonsaiDCW applied 5 to the third parameter of the constructor, colls_bits. BonsaiPR applied 6 and 8 to width_1st in 0.8 and 0.9 load factors, respectively.
The following table lists the timings needed in inserting keys (Insert), searching the keys (Search), and the maximum resident set size occupied by the insertion process (RSS). The order of keys was random.
| Data Structure | RSS (MiB) | Insert (us / key) | Search (us / key) |
|---|---|---|---|
| BonsaiDCW (0.8) | 22.4 | 1.78 | 2.67 |
| BonsaiDCW (0.9) | 20.2 | 2.05 | 5.28 |
| BonsaiPR (0.8) | 21.7 | 1.31 | 1.25 |
| BonsaiPR (0.9) | 21.6 | 1.36 | 1.30 |