TaffyDB benchmark
This benchmark is used to compare NeDB to TaffyDB 2.7 (latest version). I use the same code I use to benchmark NeDB, with the necessary adaptations. Also, since server-side TaffyDB doesn't support persistence, I benchmark it against an in-memory-only NeDB (-m options in the NeDB benchmarks), which is why the NeDB numbers for inserts, updates and removes are higher here than on the NeDB readme.
You need to npm install the dependencies to run the benchmarks. Use the -n option to specify dataset size (default: 10,000), for example node benchmarks/find.js -n 5000 will run the benchmark with a collection of 5000 documents.
Results
NeDB is much faster than TaffyDB on all operations except insert:
- Insert: TaffyDB 15,900 ops/s VS NeDB 15,300 ops/s
- Find: TaffyDB 82 ops/s VS NeDB 41,300 ops/s
- Update: TaffyDB 84 ops/s VS NeDB 8,800 ops/s
- Remove: TaffyDB 54 ops/s VS NeDB 18,700 ops/s
Interpretation
We see that insert speeds are comparable but TaffyDB is much slower on operations that require to find documents (find, update, remove). That's because TaffyDB doesn't use indexing while NeDB does (using self-balancing binary search trees). In exchange for slightly slower insertions (very small impact), we get much faster lookups. When NeDB is used in-memory and without indexes, it is comparable to TaffyDB (only about 10% faster).
Another benefit of indexing is that it can be used at no cost to enfore a unique constraint on a field. In NeDB, unique indexes are exactly as fast as normal indexes so there is no performance drop. With Taffy you would need to scan the entire collection on every insert and update, so the performance would drop a bit on updates and tremendously on inserts (I didn't run any benchmark but my guess is that inserts would become about as slow as updates).