simtune.py

from main import *
from surrogate.surrogate import *
import random

def perturb_params(params):
	for i in range(len(params)):
		for j in range(len(params[i])):
			params[i][j] = max(0, params[i][j] * (1 + (random.random() - 0.5) / 10))
	return params	

def runModel(model, steps = NUM_SIM_STEPS, dirname = 'real', params = None):
	global opts; opts.model = model
	# Initialize Environment
	datacenter, workload, scheduler, env, stats = initalizeEnvironment(opts.env, int(opts.type), opts.model)
	if dirname == 'sim':
		env.updateParams(params)

	# Execute steps
	for step in range(steps):
		print(color.GREEN+"Execution Interval:", step, color.ENDC)
		stepSimulation(workload, scheduler, env, stats)
		if 'surrogate' in dirname and step % 10 == 0:
			params = env.generateParams()
			params = perturb_params(params)
			env.updateParams(params)

	# Cleanup and save results
	if env.__class__.__name__ == 'Serverless': datacenter.cleanup()
	return stats

def generateRandomTrace(env, dirname, steps = NUM_SIM_STEPS, params = None):
	# return saved stats if this is a prerun host set
	global opts
	opts.env = env
	os.makedirs(f"data/{dirname}", exist_ok=True)
	stats = runModel('Random', steps, dirname, params)
	stats.generateSimpleHostDatasetWithInterval(f'data/{dirname}/', 'cpu')
	stats.generateSimpleMetricsDatasetWithInterval(f'data/{dirname}/', 'avgresponsetime')
	stats.generateSimpleMetricsDatasetWithInterval(f'data/{dirname}/', 'energytotalinterval')
	if 'surrogate' in dirname:
		stats.generateSimulatorParamsWithInterval(f'data/{dirname}/')
	return stats

if __name__ == '__main__':
	global opts
	dirname = "logs/SimTune"
	if not os.path.exists("logs"): os.mkdir("logs")
	if os.path.exists(dirname): shutil.rmtree(dirname, ignore_errors=True)
	os.mkdir(dirname)
	opts.type, opts.env = 2, 'Azure'

	# Generate trace from Random scheduler
	realTrace = generateRandomTrace('Azure', 'real')

	# Train a surrogate model with generated trace
	simSurTrace = generateRandomTrace('Sim', 'sim_surrogate', NUM_SIM_STEPS * 2)
	surrogate = trainModel(HOSTS, FEATS)
	
	# Update simulator parameters
	tunedParams = opt(surrogate, HOSTS)

	# Generate more data trace using simulator
	simTrace = generateRandomTrace('Sim', 'sim', NUM_SIM_STEPS * 5, tunedParams)

	# Train scheduler using more vertical data
	tunedModel = train_scheduler(HOSTS)

	# Run trained scheduler using tuned scheduler
	opts.env, opts.model = 'Azure', 'GOBI'
	datacenter, workload, scheduler, env, stats = initalizeEnvironment(opts.env, int(opts.type), opts.model)
	scheduler.model = tunedModel

	for step in range(NUM_SIM_STEPS):
		print(color.GREEN+"Execution Interval:", step, color.ENDC)
		stepSimulation(workload, scheduler, env, stats)

	datacenter.cleanup()
	saveStats(stats, dirname)