Storage slowing down execution

Hi,

I have some troubles to make Optuna runs in parallel, and even in sequential but with a local storage. My test code is really simple:

import optuna

def objectif(trial):
    x = trial.suggest_float("x", -10., 10.)
    y = trial.suggest_float("y", -10., 10.)

    return (x - 5)**2 + (y + 4)**2

study = optuna.create_study(direction = "minimize", study_name = "TestOptuna",\
                            load_if_exists = True)
study.optimize(objectif, n_trials = 1000)

Executing the previous code takes : 8,83s. Now if I add some local storage with :

study = optuna.create_study(direction = "minimize", study_name = "TestOptuna",\
                            load_if_exists = True, storage = "mysql://root:#password@localhost/TestOptuna")

The execution takes : 2min6,6s. Same problem with sqlite.

Any idea from where the problem is coming from? Thank you for helping me!

• Optuna version: 2.9.1
• Python version: 3.8.11
• OS: Linux-5.11.0-38-generic-x86_64-with-glibc2.17

[not522]

Here is the speed benchmark result. https://github.com/optuna/optuna/runs/4062350310?check_suite_focus=true#step:6:17 The slowing down ratio (inmemory, random, 1000 v.s. sqlite, random, 1000) is about x50 in the GitHub Actions’ environment.

For future development, please let me know your request? 1000 trials in 1 second?

[nzw0301]

Thank you for trying the code! I hope the optimisation with Redis performs not too slowly.