Toward a smart IoT services placement in a Fog computing infrastructure

Directed by:  P.Stolf (IRIT­SEPIA)  J­M Pierson(IRIT­SEPIA)  T.Monteil(LAAS­SARA)

Phd student:  T. Djemai

Plan

✔Context and Challenges

✔Objectives✔Actual Work✔State of progress ✔Future Prospect

Context and challenges 

+

Fog computing paradigm

IoT environnement

Cloud paradigm

Extension

Future prospectState of progressActual WorkObjectivesContext & Challenges

➢ IoT objects proliferation 

➢ Real time, Network greedy applications. 

➢ Users  Quality of Service requirements. 

➢  Centralized distant computing infrastructure.(cloud paradigm) 

➢ Dedicated network  equipments

1

Context and challenges (2)

➢ Multi-layer Heterogeneity (Users, Network and Cloud).➢ Highly distributed geographical area.➢ High number of users, equipments (Scalability) and high mobility.➢ Quality of Service Requirements (Augmented Reality,

Connected Vehicles, Health Care).➢ Consequent energy consumption (Cloud + Network + IoT Objects).➢ Management of Fog-cloud and Fog-User communication.

Fog computing infrastructure NIST view (NIST 2017 [5]) Simplified view of Fog infrastrucutre

Future prospectState of progressActual WorkObjectivesContext & Challenges

2

Objectives

Apps requests

Orchestrator(Scheduler)

G-Network control system

G-Energy control/mangement system

Zone 1(Fog Cloudlet 1)

Zone N(Fog Cloudlet N)

Prediction system

ModelsOptimizationpolicies

IoT devices Layer

Apps Submission

     Z­Energy control/mangement system

     Z­Network control system

Z-Orchestrator

G-compute control system

➢ Model and implementation of an Autonomous Framework for IoT services placement and orchestration

Z­compute control system

Future prospectState of progressActual WorkObjectivesContext & Challenges

3

A high level view of IoT services orchestrator

Actual objectives

Future prospectState of progressActual WorkObjectivesContext & Challenges

Energy consumption minimization

➢ Energy consumed for compute.

➢ Energy consumed for the network communication. 

Timeliness and Service Deliver➢ Each service has a maximum execution time that should 

not be exceeded.➢ Each pair of services that exchange  data has a maximum 

communication time not ot be exceeded.

4

Approach and methodology(1)

I.Model

SensorSensorS1S1

S2S2ActuatorActuator

s3s3

S4S4

Future prospectState of progressActual WorkObjectivesContext & Challenges

Fog Infrastructure

A three layered hierarchical infrastructure (Cloud, Fog/edge, IoT devices) modeled by A Directed Acyclic graph

➢ Nodes = Physical equipements.➢ Vertices = Physical links.

IoT application A Directed Acyclic graph (DAG)

➢ Nodes = Services➢ Vertices= Represent data dependecies between

services.

5

Approach and methodology(2)

II.iFogSimFog environment simulator based on CloudSim.

Future prospectState of progressActual WorkObjectivesContext & Challenges

6

A view of the fog infrastructure through iFogSim GUI IoT services placement process

dp(Si,Sj) : Dependency degree between two services Si and Sj.➢ Data size exchanged between Si and Sj➢ Send frequency 

Approach and methodology(3)III.AlgorithmFirst greedy heuristic’s Pseudo algorithm (using dependency degree)

Future prospectState of progressActual WorkObjectivesContext & Challenges

IN : ­List of Infrastructure’s nodes in ascending order of capacity.        ­List of applicaiton Edges in descending order of their dependency degree.OUT : Placement strategy list {« node, {Services} » }

1. For each application’s Edges list EL do2. while all services in EL are not placed do

3.chek if si and sj are not placed4.For each node nk in nodes list NL do  5.For each node nl in nodes list NL do

  6.Check if nk & nl ressources are respectively enough for si, sj       & delays constraints are respected.

7.if E is minimal then placed si in nk and sj in nl8. if si is placed then try to place sj with E minimal ( redo 5 & 6 for  nl)

 9. if sj is placed  then try to place si with E minimal ( redo 5 & 6 for  nl)

7

F2 F1 F3 F4 NLC1 C2

(S2,S3) (S1,S4) (S4,S5) EL(S1,S2)

Each particle (solution) has :– Position X(k)

– Velocity V(k)

– pbest, gBest

– Neighbours

Particule Swarm Optimization (Principal)

A Particle

pi

pj

Particle’s motion equations :

8

Particule Swarm Optimization(Base Algorithm)

● A Particule is a placement vector.

● Full neighbours topology gBest=pBest.● Stop criteria is by number of iterations.

9

F1 F1 F1 C2 C1 C2

S0 S5S1 S2 S3 S4

Avantage/disadvantages

● Stochastic approach.● Distributed approach that could be suitable for large

scale topologies in fog computing. ● Parameters.

Evaluation metrics ● Energy consumption ( IoT,Fog and Cloud layers).● Average execution time of services . ● Network usage.

10

IV.Real test infrastructure 

Approach and methodology(4)

Future prospectState of progressActual WorkObjectivesContext & Challenges

11

Architecture of our realistic testbed

State of progress

➢ Placement strategies in ifogSim.➢ Random with threahsold, AS, PSO, GME,OPT

➢ Compare with Fog Only, Cloud Only and EdgeWare strategies. 

➢ CiscoIR829 Smart router manipulation.

Future prospectState of progressActual WorkObjectivesContext & Challenges

12

Future Prospects

➢ Integrate user mobility.

➢ Integrate IoT application classes according to their QoS requirements.

➢ Establish probabilistic models for resource estimation.

➢ Dynamic adaptation to context change (eg : Network congestion point, network and ressources states).

Future prospectState of progressActual WorkObjectivesContext & Challenges

13

THANK YOU FOR YOUR ATTENTION

References

[1] Z. A. Bonomi, Milito. Fog computing and its role in the internet of things.MCC’12, August 17, 2012, Helsinki, Finland, -1, 2012.

[2] C. company. Cisco fog computing with iox.IEA 4E EDNA, Technology and Energy Assessment Report, -1, 2014.

[3] L. L. Giang, Blackstock. Developing iot applications in the fog: a distributed dataflow approach.5th International Conference on the Internet of Things (IoT), -1, 2015.

[4] G. B. Gupta, Dastjerdi. ifogsim: A toolkit for modeling and simulation of resource management techniques in the internet of things, edge and fog computing environments.IEEE, -1, 2016. to appear.

[5] B. M. G. M. Iorga, Feldman. Fog computing conceptual model recommendations of the national institute of standards and technology.NIST Special Publication 500-325, -1, 2017.