You are on page 1of 1

L IST OF F IGURES

1.1 Work assignment and processing: schematic demonstration. . . . . . . . . . . 4

2.1 Schematic view of all four proposed string matching approaches. . . . . . . . . 20


2.2 Summary of results: Running time vs. pattern size. . . . . . . . . . . . . . . . 23
2.3 Running time vs. pattern and text length, and superior algorithms vs. parameters. 30
2.4 Running time vs. pattern size, with a fixed text size. . . . . . . . . . . . . . . . 33

3.1 Multisplit examples. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47


3.2 An example for our localization terminology. . . . . . . . . . . . . . . . . . . 54
3.3 Hierarchical localization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.4 Different localizations for DMS, WMS and BMS are shown schematically. . . 62
3.5 Key distributions for different multisplit methods and different number of buckets. 71
3.6 Running time vs. number of buckets for all proposed methods. . . . . . . . . . 82
3.7 Achieved speedup against regular radix sort. . . . . . . . . . . . . . . . . . . . 83
3.8 Running time vs. number of buckets for various input distributions. . . . . . . . 89

4.1 Insertion example in GPU LSM. . . . . . . . . . . . . . . . . . . . . . . . . . 111


4.2 Batch insertion time, and effective insertion rate for GPU LSM. . . . . . . . . 124

5.1 Regular linked list and the slab list. . . . . . . . . . . . . . . . . . . . . . . . . 133


5.2 Psuedo-code for search (SEARCH) and insert (REPLACE) operations in slab hash. 140
5.3 Memory layout for SlabAlloc. . . . . . . . . . . . . . . . . . . . . . . . . . . 142
5.4 Performance the slab hash and cuckoo hashing versus memory efficiency. . . . 145
5.5 Performance versus total number of stored elements. . . . . . . . . . . . . . . 147
5.6 Incremental batch update for the slab hash, and bulk build for the cuckoo hashing. 148
5.7 Concurrent benchmark for the slab hash. . . . . . . . . . . . . . . . . . . . . . 150

-viii-

You might also like