A second benchmark was conducted using an Intel N100 CPU to evaluate performance on lower-power devices.
Exact Search Time (ms) (English)
| Text Size | Linear | GPU | Bigram |
|---|
| 538 | 1.04 | 1.02 | 0.20 |
| 829 | 1.47 | 1.55 | 0.28 |
| 1712 | 2.42 | 2.54 | 0.24 |
| 3001 | 4.26 | 4.32 | 0.37 |
| 3748 | 5.21 | 4.98 | 0.46 |
Fuzzy Search Time (ms) (English)
| Text Size | Linear | GPU | Bigram |
|---|
| 538 | 21.14 | 8.24 | 0.56 |
| 829 | 31.67 | 10.06 | 0.88 |
| 1712 | 65.76 | 14.93 | 0.98 |
| 3001 | 115.56 | 20.40 | 1.16 |
| 3748 | 142.11 | 25.72 | 1.32 |
Fuzzy Search Time (ms) (Japanese)
| Text Size | Linear | GPU | Bigram |
|---|
| 476 | 7.60 | 7.45 | 0.32 |
| 789 | 10.70 | 6.67 | 0.45 |
| 1305 | 17.43 | 7.25 | 0.44 |
| 2394 | 31.17 | 8.20 | 0.57 |
| 3020 | 38.89 | 9.06 | 0.75 |
Indexing Time (ms) (English)
| Text Size | Linear | GPU | Bigram |
|---|
| 538 | 126.14 | 117.40 | 261.80 |
| 829 | 188.64 | 185.20 | 406.52 |
| 1712 | 346.98 | 362.18 | 862.26 |
| 3001 | 592.98 | 617.08 | 1,686.10 |
| 3748 | 744.78 | 737.06 | 1,892.40 |
Index Size (Gzipped, kbyte) (English)
| Text Size | Linear | GPU | Bigram |
|---|
| 538 | 191 | 191 | 91 |
| 829 | 294 | 294 | 131 |
| 1712 | 607 | 607 | 247 |
| 3001 | 1058 | 1058 | 411 |
| 3748 | 1324 | 1324 | 506 |
Index Size (Gzipped, kbyte) (Japanese)
| Text Size | Linear | GPU | Bigram |
|---|
| 476 | 165 | 165 | 136 |
| 789 | 272 | 272 | 229 |
| 1305 | 452 | 452 | 366 |
| 2394 | 837 | 837 | 644 |
| 3020 | 1053 | 1053 | 813 |
Conclusion
In summary, the benchmarks conducted on the Intel N100 CPU demonstrate the performance characteristics of different indexing strategies. While the LinearIndex provides reliable performance, the GPU and HybridBigramInvertedIndex indexes offer significant speed improvements, especially for larger datasets. When selecting an index type, consider the specific requirements of your application and the hardware capabilities available.
