Obviously all times are dependant on hardware, so only consider these relative to each other. There will always be slight variation but the figures shown here are pretty stable.
| Method | Real | User | Sys |
|---|---|---|---|
| sqlDumpAndExec (full) | 4m51.927s | 5m18.130s | 0m5.070s |
| sqlDumpAndExec (read to /dev/null) | 0m21.239s | 0m32.290s | 0m0.880s |
| sqlDumpAndExec (read to file) | 0m29.755s | 0m32.290s | 0m0.880s |
| sqlDumpAndExec (write from file) | 4m49.877s | 4m43.050s | 0m2.570s |
| sqlalchemy1 | gave up after 8 min | ||
| sqlalchemyBulkInsert 200 | 14m54.983s | 7m42.980s | 0m19.990s |
| sqlalchemyBulkInsert 2000 | 8m13.869s | 6m44.830s | 0m3.770s |
| sqlalchemyBulkInsert 20000 | 7m31.308s | 6m48.250s | 0m2.700s |
| sqlalchemyMemory 20000 | 7m7.145s | 6m48.540s | 0m1.560s |
| sqlalchemyPragmas 20000 | 7m8.928s | 6m49.700s | 0m2.420s |
| sqlalchemyReadOnly | 2m23.909s | 2m6.470s | 0m0.680s |
| sqlalchemyFetchAll | 7m5.355s | 6m19.440s | 0m3.340s |
| customCpp | 1m10.478s | 0m45.390s | 0m1.280s |
After these tests I got a new laptop (MacBookPro with 2.3 GHz Intel Core i5 and SSD), so the following can't be compared to the table above.
| Method | Real | User | Sys |
|---|---|---|---|
| customCpp (native) | 0m4.441s | 0m2.493s | 0m0.323s |
| customCpp (docker container) | 0m7.08s | 0m 4.04s | 0m 0.98s |
| sqlpp11 | 0m28.708s | 0m24.231s | 0m3.449s |
Interestingly the docker container adds a slight overhead; this time doesn't include spinning up the container. I suspect this is either better compiler optimisation or performance of the docker I/O driver (RAW in v17.12.0-ce). It's minimal compared to the differences from the code so I can't be bothered figuring out more details.
sqlpp11 doesn't copy all the data either. See the readme in that folder.