search

park-project / park

MIT License
218 stars 55 forks source link

About load_balance RL agent #20

Closed lesleychou closed 3 years ago

lesleychou commented 3 years ago

Hi @hongzimao,

Thanks for your well-commented work!

I am trying to reproduce the Server load balancing results from Fig.4 in the paper. image

However, after I added the A2C agent, the reward is not increasing over 1e9 epochs. I debugged my A2C agent, it learned successfully in three OpenAI gym envs.

I am wondering if it is possible for you to share the agent file you had for load balancing? If so it would be greatly helpful for us!

hongzimao commented 3 years ago

This load balancer example should be similar to one of other paper: http://people.csail.mit.edu/hongzi/var-website/index.html (see section 6.2 and figure 5). The A2C agent is open sourced in https://github.com/hongzimao/input_driven_rl_example/blob/master/load_balance_actor_agent.py hope it's helpful for your scenario.

lesleychou commented 3 years ago

Thank you! It's extremely helpful, I was able to reproduce the paper results. Sincerely appreciate your active response as always!