nlp-database-read-performance
Every test was preceded by restarting server (in case of Mongo) and dropping disk buffer. All tests were performed using Ruby 1.9.2 (if not specified differently).
sync ; sudo sh -c 'echo 3 > /proc/sys/vm/drop_caches'
sudo service mongodb restart
Tests:
- All documents - retrieve all documents (~ million)
- 10000 ID documents - find randomly generated 10000 IDs with the same seed; 9732 IDs were not present in database
Above tests were also performed with 10 times repetition. It tests cache systems and use of disk buffer.
| All documents | 10000 ID documents | 10 times all documents | 10 times 10000 ID documents | |
|---|---|---|---|---|
| Rod | 0m25.167s | 0m30.412s | 3m15.944s | 0m36.030s |
| MongoMapper 0.12 (IdentityMap) | 2m36.715s | 0m14.351s | ||
| MongoMapper 0.13b | 0m51.931s | 0m12.359s | 7m7.040s | 0m48.353s |
| MongoMapper 0.13b (IdentityMap) | 0m47.529s | 0m8.363s | 7m17.493s | 0m44.588s |
| MongoMapper 0.13b (Ruby 2.1) | 0m50.607s | 0m10.465s | 1m7.621s | |
| MongoMapper 0.13b (Ruby 2.1, IdentityMap) | 0m47.085s | 0m9.714s | 1m3.656s | |
| Mongo | 0m33.139s | 0m5.793s | 3m57.887s | 0m42.071s |
| Mongo (Ruby 2.1) | 0m25.711s | 0m6.143s | 4m4.164s | 0m36.973s |
| Mongoid (Ruby 1.9.3) | 1m10.046s | |||
| Mongoid (identity map, Ruby 1.9.3) | 1m46.000s | |||
| Mongoid (Ruby 2.1) | 0m55.921s | 0m6.906s | 0m51.628s | |
| Mongoid (identity map, Ruby 2.1) | 0m45.754s |
| BFS (14000 nodes) | |
|---|---|
| Rod | 2m41.914s |
| MongoMapper 0.13b | 0m12.941s |
- Tests should be measured in code.
- Dropping disk buffer command is questionable.
- Documents in Mongo are smaller (less fields).
Mongo performance is comparable to Rod. Mongo is better in random access or indexing. ORM for Mongo can be worse twice times than Rod.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent