Park

Real system interface

import park
import agent_impl  # implemented by user

env = park.make('congestion_control')

# the run script will start the real system
# and periodically invoke agent.get_action
agent = Agent(env.observation_space, env.action_space)
env.run(agent)

The agent_impl.py should implement

class Agent(object):
    def __init__(self, state_space, action_space, *args, **kwargs):
        self.state_space = state_space
        self.action_space = action_space

    def get_action(self, obs, prev_reward, prev_done, prev_info):
        act = self.action_space.sample()
        # implement real action logic here
        return act

Simulation interface

Similar to OpenAI Gym interface.

import park

env = park.make('load_balance')

obs = env.reset()
done = False

while not done:
    # act = agent.get_action(obs)
    act = env.action_space.sample()
    obs, reward, done, info = env.step(act)

Contributors

Environment	env_id	Committers
Adaptive video streaming	abr, abr_sim	Hongzi Mao, Akshay Narayan
Spark cluster job scheduling	spark, spark_sim	Hongzi Mao, Malte Schwarzkopf
SQL database query optimization	query_optimizer	Parimarjan Negi
Network congestion control	congestion_control	Akshay Narayan, Frank Cangialosi
Network active queue management	aqm	Mehrdad Khani, Songtao He
Tensorflow device placement	tf_placement, tf_placement_sim	Ravichandra Addanki
Circuit design	circuit_design	Hanrui Wang, Jiacheng Yang
CDN memory caching	cache	Haonan Wang, Wei-Hung Weng
Multi-dim database indexing	multi_dim_index	Vikram Nathan
Account region assignment	region_assignment	Ryan Marcus
Server load balancing	load_balance	Hongzi Mao
Switch scheduling	switch_scheduling	Ravichandra Addanki, Hongzi Mao

Misc

Note: to use argparse that is compatiable with park parameters, add parameters using

from park.param import parser
parser.add_argument('--new_parameter')
config = parser.parse_args()
print(config.new_parameter)