POWER-AWARE MULTI-DATACENTER MANAGEMENT USING MACHINE LEARNING

CLOUDCOMPUTING

Presented by: Omar Sulca

CONTENT

1. Introduction

2. What they looking for?

3. What is Multi Data Center?

4. Managing Multi DCs

5. Modeling the System

6. Conclusions

1. INTRODUCTION

Cloud Computing, has become crucial for the externalization of IT resources for business, organizations and people.

“everything as a service”

(plataform, infrastructure and service)

Providers want in turn to optimize the use of the resources they have deployed with their own metrics

1. INTRODUCTION

Factors to be optimized

Consolidation - Set the maximum number of services in the least viable amount of hosting machines, so the number of on-line machines and resources is minimized.

Virtualization technology has made consolidation easier,

Revenues• came from servicing the clients of the hosted

web-services with reasonable Quality of Service (QoS)

Costs • operational costs for the infrastructure (Energy-realeted cost)

2. WHAT THEY LOOKING FOR?

“Build a model to automate (AC) an improve the process of achieveallocation of virtualized web-services, using a Machine Learning (ML)

and Data Mining, to predict behavior and select “policies” to be appliedin a multi-DC”

Energy Saving in Cloud Self-management

3. WHAT IS MULTI DATA CENTER?

• Its a Networking of Data Centers (DCs) interconected

Must be considerate

Migration overheads

Service-Client proximity

Energy cost at diferent locations

Modularity between inter-DC relations aninformation

4. MANAGING MULTI DCS

Multi-DataCenter Business Model

1

2

SLA (Service Level Agreement)3

ensure the agreed QoS forde VM, while minimizing the

cost by reducing theresorces usage4

5. MODELING THE SYSTEM

In this case

Quality of Service = Response Time

Mathematical Model(monitoring PM resources

and adjusting VM placementsand quotas)

Machine Learning + Data Mining

Predict behavior andScheduling theVM

Across de DC networks

Using

to

Around the world

5. MODELING THE SYSTEM

The Machine Learning decisivefactors

Energy consumption

Resource allocation

QoS

Questions to predict

How good will each VM behave?

How much CPU/Mem/IO… will eachVM demand?

6. CONCLUSIONS

1. Optimizing the schedule and management of multi-DC systems requires balancing several factors, like economic revenues, Quality of Service and operational costs such as energy.

2. Using virtualization technology is presented a model to solve a multi-DC scheduling problem which balances and optimizes the economic factors above.

3. A few issues for future study are:

a) How decide which VMs are excluded from inter-DC scheduling or which PMs are offered as host candidates for scheduling;

b) The inclusion of more operational costs (networking, bandwidth management,etc.)

c) The green energy into the scheme and the environmental impact of computation.

d) The use of online learning methods to make the system react quickly to changes (application behavior, hardware or middleware changes, or workload characteristics

Thanks