ServlessSimPro

A Comprehensive Serverless Simulation Platform in a Cloud Environment.

Reference article:

ServlessSimPro: A comprehensive serverless simulation platform

Download and Run

When downloading the code, please use Git.

After git clone, make sure you already download /Trace/AzureFunctionsInvocationTraceForTwoWeeksJan2021.txt. This is the default dataset of our simulator.

It is recommended to use Python 3.9 as the Python interpreter version.

You can start the simulation by running simulator.py.

Project Structure

• EnumClass helps to manage enum class in the system.
• Logs directory contains running logs, results, and plotting functionality.
• Resource directory contains different resource classes and request classes used in the system.
• SchedulingInterface directory contains different scheduling algorithm classes.
• Trace directory contains the dataset used in this project.
• Simulator.py is the main program for serverless simulation.

Parameter Adjustment

You can adjust the simulation parameter in simulator.py:

from resource.container import Container
from resource.pm import PM, CPU
from resource.app import App, APP_NUM
import resource.req
import resource.app
from schedulingInterface.ContainerPlacement import FirstFit, MinVectorDist
from schedulingInterface.RequestAllocation import EarliestKilled, LatestKilled, RandomSelection
from schedulingInterface.ContainerConsolidation import MinPmNum
from schedulingInterface.PopQueue import FCFS, SJF, HRRN
from enumClass.enumClass import ReqAllocAlgo, ConPlaceAlgo, ConConsAlgo, PopQueueAlgo, Task, ContainerState

# ####################### Simulation Parameters(Adjustable) #######################
req_num = 1000
P_idle = 92.61
P_max = 259.67
P_mid = 94.8
cs_factor = 0.52

reuseTimeWindow = 600

USE_CONSOLIDATION = True
CONSOLIDATION_TIME_INTERVAL = 300
CONSOLIDATION_THRESHOLD = 0.38

LOG_TIME_INTERVAL = 50

APP_CONF_RANDOM = False

USE_QUEUE = True
PARAM1 = 50
PARAM2 = 50
QUEUE_THRESHOLD = [PARAM1] * APP_NUM
MAX_QUEUE_LENGTH = [PARAM2] * APP_NUM

# ####################### Strategies(Adjustable) #######################

ContainerPlacementStrategy = ConPlaceAlgo.FIRST_FIT
RequestAllocationStrategy = ReqAllocAlgo.EARLIEST_KILLED
ContainerConsolidationStrategy = ConConsAlgo.MIN_PM_NUM
PopQueueStrategy = PopQueueAlgo.FCFS

# log file name in /logs/plotFigs
LOG_STRATEGY = RequestAllocationStrategy

For Simulation Parameters(Adjustable):

• req_num is the number of arrival requests.
• P_idle, P_max, P_mid, cs_factor are related to power of the physical machine.
• reuseTimeWindow is the reuse time of the container.
• USE_CONSOLIDATION, CONSOLIDATION_TIME_INTERVAL, CONSOLIDATION_THRESHOLD are related to container consolidation strategies.
• LOG_TIME_INTERVAL is the time interval of logging.
• APP_CONF_RANDOM is whether to use random application configuration.
• USE_QUEUE, QUEUE_THRESHOLD, MAX_QUEUE_LENGTH are related to queuing strategies.

For Strategies(Adjustable): refer to Strategies Readme

Noted: Global Variables are simulation variables, their values change during simulation.

Simulation Results

Metrics

After simulation, the results can be found in logMetrics. We provide some metrics such as:

Total Energy Consumption
Total Latency
Cold Start Count
Rejection Count
Average CPU Allocation Rate
Average memory Allocation Rate
Average Maximum Concurrency
Average Cold Start State Count
Average Running State Count
Average Idle State Count
Average Dead State Count
Average Active Physical Machine Count

If you want more customized metrics, add them inside the logInfo function in the simulator.py.

Plot Figs

We provide plotting functionality. If you want to plot system metrics, please refer to the Plot Readme.

Contact Information

你好 Hello Hola!

Welcome to use this platform for your research work, and feel free to continue improving this simulation platform. If you have any questions, please contact 220222036@seu.edu.cn.