PhD Defence

Dec 8, 2017 - Virtual Machine Allocation in Fog Computing : ▻ Multiple resources are needed : RAM, CPU, Bandwidth, etc... ▻ Many types VM requests : ↿⇂ RAM or ↿⇂ CPU... ▻ Processing facilities with “finite” capacity of each resource. ▻ Centralised policies are not suitable. G. Thompson. PhD Defence: INRIA Paris, ...
894KB Sizes 1 Downloads 107 Views
Stochastic Models for Resource Allocation in Large Distributed Systems Guilherme THOMPSON Advisor: Philippe ROBERT

Paris,

INRIA Paris of December 2017

08th

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

1/40

Agenda

1. Introduction 2. Resource Allocation with Downgrading 3. Multi-resource Cooperation in the framework of Cloud Computing 4. Conclusion

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

2/40

Introduction Stochastic Modelling of Large Distributed Systems

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

3/40

Introduction Stochastic Modelling of Large Distributed Systems Cloud Computing and its new architectures

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

3/40

Introduction Stochastic Modelling of Large Distributed Systems Cloud Computing and its new architectures Decentralisation of data centres — smaller processing facilities — closer to end-user

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

3/40

Introduction Stochastic Modelling of Large Distributed Systems Cloud Computing and its new architectures Decentralisation of data centres — smaller processing facilities — closer to end-user

Fog Computing, VoD, IoT, VNF...

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

3/40

Introduction . Motivation Decentralisation PM 1 VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VM

PM 2

PM N

DC

N Physical Machines

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

4/40

Introduction . Motivation Decentralisation

PM 1

PM 1

PM N1

PM 2

DC 1

VM VMVM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VMVM VM VM VM VM VM VM VMVM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VMVM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VM VMVM VM VM VM VM VM VM

PM N2

PM 2

DC 2

N1 + N2 = N Physical Machines G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

4/40

Introduction . Motivation Stochastic Modelling of Large Distributed Systems

Distributing processing facilities results in — lower: costs, delays, energy consumption... — higher: load variability, mismatch of load and capacity

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

5/40

Introduction . Motivation Stochastic Modelling of Large Distributed Systems

Distributing processing facilities results in — lower: costs, delays, energy consumption... — higher: load variability, mismatch of load and capacity Congestion Control Mechanisms — Cooperation among processing facilities — Adaptation to incoming requests

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

5/40

Introduction . Motivation Stochastic Modelling of Large Distributed Systems Resource allocation in Cloud Computing I I I I

static : load balancing, knapsack problems... resource sharing (fairness): DRF, BF... scheduling : PoC, redundancy... others : energy, network use...

G. Thompson

PhD Defence: INRIA Paris, 08/12/2017

6/40

Introduction . Motivation Stochastic Modelling of Large Distributed Systems Resource allocation in Cloud Computing I I I I

static : load balanc