Measuring the Fog, Gently Antonio Brogi, Stefano Forti, and Marco - - PowerPoint PPT Presentation

β–Ά
measuring the fog gently
SMART_READER_LITE
LIVE PREVIEW

Measuring the Fog, Gently Antonio Brogi, Stefano Forti, and Marco - - PowerPoint PPT Presentation

Aug 13, 2022 185 likes β€’417 views

Measuring the Fog, Gently Antonio Brogi, Stefano Forti, and Marco Gaglianese Service-oriented, Cloud and Fog Computing Research Group Department of Computer Science University of Pisa, Italy 17th International Conference on Service-Oriented

slide-1
SLIDE 1

Measuring the Fog, Gently

Antonio Brogi, Stefano Forti, and Marco Gaglianese

Service-oriented, Cloud and Fog Computing Research Group Department of Computer Science University of Pisa, Italy

17th International Conference on Service-Oriented Computing, October 28-31, 2019, Toulouse, France

slide-2
SLIDE 2

Embedded AI Autonomous driving Drones for deliveries Energy production Smart Cities

2

CONTINUOUS IOT GROWTH

slide-3
SLIDE 3

3

microservices multi-component

  • smotic

LARGE HIGHLY DISTRIBUTED SOFTWARE SYSTEMS

slide-4
SLIDE 4

4

mist cloud micro-cloud fog IoT edge

PERVASIVELY DISTRIBUTED INFRASTRUCTURES

slide-5
SLIDE 5

5

STRINGENT QoS REQUIREMENTS

slide-6
SLIDE 6

How to adaptively manage

LARGE HIGHLY DISTRIBUTED SOFTWARE SYSTEMS

  • n top of

PERVASIVELY DISTRIBUTED INFRASTRUCTURES

so so to

  • guarantee their

STRINGENT QoS REQUIREMENTS

?

6

slide-7
SLIDE 7

Fog

  • g Orch

chestratio ion La Layer

  • First proposed by Bonomi et al. in 2014
  • Since then:
  • Much work* to cover Analyse
  • Some work for Plan and Execute
  • Few work on Monitor
  • Monitor is important to decide:

1. where to deploy app services at first 2. when/where to migrate app services

* A. Brogi, S. Forti, C. Guerrero, I. Lera. "How to Place Your Apps in the Fog-State of the Art and Open Challenges." Software Practice and Experience (In Press)

slide-8
SLIDE 8

Chall llenges in in Fog

  • g Mon
  • nit

itorin ing

unstable Internet connectivity heterogeneous & resource-constrained devices failures and churn

slide-9
SLIDE 9

Rela lated Wor

  • rk: Fog
  • g Mon
  • nit

itorin ing

To the best of our knowledge, none target all three challenges.

slide-10
SLIDE 10

Our Proposal

A lightweight fault-resilient monitoring technique for Fog infrastructures, prototyped in an open-source tool

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

slide-11
SLIDE 11

metr tric ics

The prototype monitors:

  • Hardware resources availability
  • CPU, RAM, HDD
  • End-to-end (e2e) network QoS
  • latency and bandwidth
  • Connected IoT devices
slide-12
SLIDE 12

ar arch chit itectu ture

Two types of distributed P2P agents:

  • Followers measuring

monitored metrics, and

  • Leaders aggregating

metrics from a group of Followers, and gossiping them to other Leaders

slide-13
SLIDE 13

Foll llower

Latency-aware Leader selection

slide-14
SLIDE 14

Le Leader

  • Leaders collect measurements from Followers in

their groups.

  • Leaders spread data to other Leaders via

Gossiping

  • 𝑃(log 𝑀) rounds to spread information on avg
  • 𝑃(𝑀 log 𝑀) messages exchanged overall
slide-15
SLIDE 15

Meas asurin ing Har ardware an and IoT IoT

  • Hardware capabilities (CPU,RAM,HDD)
  • Hyperic Sigar API
  • IoT devices connected via serial port (USB & Bluetooth)
  • libserialport tool (iThing and iIoTDiscoverer)
slide-16
SLIDE 16

Meas asurin ing La Latency

ICMP via ping for intra-group and leader-to-leader measurements Inter-groups ℓ𝐡,𝐢 ≃ ℓ𝐡,𝑀1 + ℓ𝑀1,𝑀2 + ℓ𝑀2,𝐢 assuming Leader-Leader latency is higher than Leader-Follower latency

slide-17
SLIDE 17

Meas asurin ing Ban andwidth

  • Intrusive measurements
  • Iperf3

T1 T2 T2 Congestion

  • Passive techniques
  • Assolo

𝛾𝐡,𝐢 ≃ min

𝑙,β„Ž max 𝛾𝐡,𝑙 , max 𝛾𝐢,β„Ž

Inter-groups:

slide-18
SLIDE 18

Fau ault lt-tolerance an and Sc Scala labil ilit ity

Fault-tolerance

  • Data replication at Leaders

guarantees tolerance wrt some Leader failures.

  • Followers rearrange into other

groups when their Leader fails.

  • Groups keep working in case of

network interruption between Leaders.

Scalability

  • 𝑂 nodes, 𝑀 leaders
  • 𝑂

𝑀 nodes per group (per Leader)

  • 𝑂2

𝑀2 e2e measurements for bw

and latency

  • If 𝑀 ≃

𝑂 then 𝑃

𝑂2 𝑂

2

= 𝑃(𝑂) e2e measurements.

slide-19
SLIDE 19

Testb tbed

  • 12 heterogeneous

nodes

  • Heterogenous Internet

Access:

  • ADSL 7/1,
  • VDSL 70/20,
  • VDSL 20/3,
  • FiberLAN
  • micro:bit IoT devices
slide-20
SLIDE 20

Performance

<2% CPU <3 MB of RAM <5% bandwidth usage

  • Measurement error intra-group:
  • Latency & bandwidth bound by 5%
  • Avg measurement error inter-group:
  • Latency: 14%
  • Bandwidth: 18%
slide-21
SLIDE 21

Co Conclusio ions & & Futu ture Wor

  • rk

Fog infrastructure monitoring techniques Open-source prototype Lightweight, fault- tolerant and scalable Leader election mechanisms Improve network QoS estimates further experimental assessment

slide-22
SLIDE 22

Measuring the Fog, Gently

Antonio Brogi, Stefano Forti and Marco Gaglianese

Service-oriented, Cloud and Fog Computing Research Group Department of Computer Science University of Pisa, Italy 17th International Conference on Service-Oriented Computing, October 28-31, 2019, Toulouse, France https://github.com/di-unipi-socc/FogMon