How to best deploy your Fog applications, probably Stefano Forti - - PowerPoint PPT Presentation

how to best deploy your fog applications probably
SMART_READER_LITE
LIVE PREVIEW

How to best deploy your Fog applications, probably Stefano Forti - - PowerPoint PPT Presentation

How to best deploy your Fog applications, probably Stefano Forti Ahmad Ibrahim Antonio Brogi 1 st International Conference on Fog and Edge Computing 2017 14 th May 2017 name.surname@di.unipi.it IoT and Cloud Computing 50 billion of connected


slide-1
SLIDE 1

How to best deploy your Fog applications, probably

Antonio Brogi

Stefano Forti Ahmad Ibrahim

1st International Conference on Fog and Edge Computing 2017 14th May 2017 name.surname@di.unipi.it

slide-2
SLIDE 2

IoT and Cloud Computing

  • The Cloud alone cannot support the IoT momentum.
  • There is a need for filtering and processing before the Cloud.

2

50 billion of connected devices by 2020

slide-3
SLIDE 3

Fog Features

3

  • App deployments dynamically adapt to the state of the network.

QoS-awareness

  • Position is known so to handle fluid and mobile computation.

Location-awareness

  • Discover and use available resources, cooperating horizontally.

Context-awareness

slide-4
SLIDE 4

Open Problems

4

  • How to automatically decide

where to deploy each component of an application by exploiting QoS-, location-, and context-awareness?

  • How to estimate the QoS-

assurance of a candidate deployment?

slide-5
SLIDE 5

Motivating example

5

DataStorage Dashboard ThingsController video water moisture fire VDSL 3G Sat.

slide-6
SLIDE 6

Concretely...

6

HOW MANY AND HOW

POWERFUL FOG NODES DO I NEED TO ADEQUATELY DEPLOY MY APPLICATION?

SHOULD I DEPLOY THIS

COMPONENT ONTO THE CLOUD, ONTO A FOG-AS-A-SERVICE OPENED IN MY CITY OR ON MY PREMISES GATEWAY?

IS THERE ANY COMPONENT I’D

BETTER DEPLOY ON A DIFFERENT NODE AFTER THIS LINK/NODE FAILURE?

slide-7
SLIDE 7

Concretely…

7

DO I HAVE TO UPGRADE

MY INFRASTRUCTURE IF THE APPLICATION REQUIREMENTS CHANGE?

IS IT POSSIBLE TO

REDUCE RESOURCE CONSUMPTION OF SOME

FOG NODES, OR AVOID

THEM?

WHICH ARE THE ELIGIBLE

DEPLOYMENTS THAT COMPLY MOST WITH THE REQUIRED QOS?

slide-8
SLIDE 8

Our Solution

8

Modelling of IoT apps and Fog infrastructures Algorithms to determine eligible deployments Evaluation of

  • utput deployments

via Monte Carlo

slide-9
SLIDE 9

Our Prototype

9

https://github.com/di-unipi-socc/FogTorchPI

slide-10
SLIDE 10

QoS Profiles

  • A QoS profile is a pair

ℓ, 𝑐↓, 𝑐↑

  • They represent latency and bandwidth featured by a

link or requested by a software interaction.

10

AVERAGE LATENCY AVERAGE DOWNLOAD & UPLOAD BANDWIDTH

slide-11
SLIDE 11

Application

11

160 𝑛𝑡, 0.5 𝑁𝑐𝑞𝑡, 0.7 𝑁𝑐𝑞𝑡 SD video

slide-12
SLIDE 12

Infrastructure

12

98% ⟨70 𝑛𝑡, 6 𝑁𝑐𝑞𝑡, 0,75 𝑁𝑐𝑞𝑡⟩ 2% 70 𝑛𝑡, 0𝑁𝑐𝑞𝑡, 0 𝑁𝑐𝑞𝑡 Satellite 7M

slide-13
SLIDE 13

Compatibility

  • A software component is compatible with a Fog or

Cloud node when its software and hardware* can support at least that component.

13

* Hardware only for Fog nodes.

✓ ✓ ✓ X

slide-14
SLIDE 14

Things Binding

  • Software components may have Things requests.
  • Each request is bound to a specific Thing before deployment.

14

ThingsController video water moisture fire

slide-15
SLIDE 15

Deployment Policy

  • A start-up sponsored by a specific Cloud provider,
  • an automated industrial plant,
  • an invoked third party service...

...may enforce legal, commercial or political constraints for deploying an application.

  • We allow specification of a whitelist of nodes permitted

for installing each component.

15

slide-16
SLIDE 16

Eligible Deployments

  • An eligible deployment for an application over

a Fog infrastructure ensures

16

Compatibility and deployment policies Hardware resources Things binding Bandwidth and latency

slide-17
SLIDE 17

NP-hard Problem*

17 [Garey, Michael R., and David S. Johnson. Computers and Intractability (1979)]

* By reduction from Subgraph Isomorphism.
  • A. Brogi and S. Forti, QoS-aware Deployment of IoT Applications Through the Fog, in IEEE Internet of Things Journal, 2017.

Backtracking strategy to explore the search space.

slide-18
SLIDE 18

Bird’s eye view

18

https://github.com/di-unipi-socc/FogTorchPI

slide-19
SLIDE 19

Monte Carlo Simulator

Repeat a sufficiently large number of times:

  • 1. Sample a QoS profile for each link in the infrastructure.
  • 2. Run backtracking algorithm.

Compute QoS-assurance of generated deployments.

19

slide-20
SLIDE 20

FogTorchΠ Results

20

WHICH ARE THE ELIGIBLE

DEPLOYMENTS THAT COMPLY MOST WITH THE REQUIRED QOS?

slide-21
SLIDE 21

FogTorchΠ Results (1)

21

IS IT POSSIBLE TO

REDUCE RESOURCE CONSUMPTION OF SOME

FOG NODES, OR AVOID

THEM?

E.g., avoid using fog_3 for deployment.

slide-22
SLIDE 22

FogTorchΠ Results (1)

22

IS IT POSSIBLE TO

REDUCE RESOURCE CONSUMPTION OF SOME

FOG NODES, OR AVOID

THEM?

E.g., avoid using fog_3 for deployment. ✓

slide-23
SLIDE 23

FogTorchΠ Results (2)

23

DO I HAVE TO UPGRADE

MY INFRASTRUCTURE IF THE APPLICATION REQUIREMENTS CHANGE?

E.g., deploying HD video streaming without upgrade, leads to same QoS-assurance.

slide-24
SLIDE 24

FogTorchΠ Results (2)

24

DO I HAVE TO UPGRADE

MY INFRASTRUCTURE IF THE APPLICATION REQUIREMENTS CHANGE?

Deploying HD video streaming without upgrade, leads to worse QoS-assurance.

X

slide-25
SLIDE 25

Results FogTorchΠ (3)

25

slide-26
SLIDE 26

Results FogTorchΠ (3)

26

slide-27
SLIDE 27

Conclusions

27

Determine, simulate and compare eligible deployments QoS- and context- awareness of deployments Evaluation of QoS variations impact based on links data

slide-28
SLIDE 28

Future Work

28

Design a cost model to improve search & evaluation Include multiple and multi-tenant deployments Assessment over case studies

slide-29
SLIDE 29

Thanks

Q&A

!

slide-30
SLIDE 30

Roles and Stakeholders