16
Controlling Cloud Services Elasticity in Heterogeneous Clouds Georgiana Copil, Daniel Moldovan, Hong-Linh Truong, Schahram Dustdar Vienna University of Technology 7 th IEEE/ACM International Conference on Utility and Cloud Computing 6 th Cloud Control Workshop

Controlling Cloud Services Elasticity in Heterogeneous Clouds - UCC 2014 - CloudControl6

Embed Size (px)

Citation preview

Controlling Cloud Services Elasticity

in Heterogeneous Clouds

Georgiana Copil, Daniel Moldovan, Hong-Linh Truong,

Schahram Dustdar

Vienna University of Technology

7th IEEE/ACM International Conference on Utility and Cloud Computing

6th Cloud Control Workshop

UCC 2014, 8-11 December, London 2

Overview

Background

Motivation

Approach

Results

Conclusions

UCC 2014, 8-11 December, London 3

Background

Control Application

Monitoring

Control

processes

Metrics

monitoring

information

Elasticity

requirements

Application

description

Aggregated

information

Elasticity

behavior

analysis

Application model: structural and runtime information

UCC 2014, 8-11 December, London 4

Motivation

Single application: the controller has an end-to-end

view of the application

Should we control it uniformly or leverage

cloud heterogeneity?

How can we take into consideration relationships among

components deployed on different cloud providers?

UCC 2014, 8-11 December, London 5

Research problems and approach

(Q) Should we control it uniformly or leverage cloud

heterogeneity?

– Heterogeneous control – understand a variety of control

primitives and protocols from a variety of providers

(Q) How can we take into consideration relationships

among components from different cloud providers?

– Model relationship types, and consider them in the control

process

UCC 2014, 8-11 December, London 6

Enforcement on heterogeneous clouds

Extend rSYBL to support heterogeneous cloud control

– Supporting multiple enforcement plugins simultaneously

– Parallelization of actions on multiple enforcement plugins

UCC 2014, 8-11 December, London 7

Application model for heterogeneous clouds

Extending the model for heterogeneous cloud application

control

UCC 2014, 8-11 December, London 8

Application model for heterogeneous clouds

Extending the infrastructure system description for

heterogeneous cloud application control

– Supporting the description of complex systems - various virtual

resources from multiple cloud providers

UCC 2014, 8-11 December, London 9

Application model for heterogeneous clouds

Elasticity relationships description for understanding control

dependencies

– Single way load dependency

• a change in the antecedent load causes

a similar change in consequent load

– Two way load dependency

• a change in one side causes a similar

change in the other

– Instantiation dependency

• for the instantiation of the consequent

the antecedent should to exist

– Data dependency

• the specified data should to be

transferred among the two

– Polynomial relationship

• a polynomial function describes the

connection between the two components

UCC 2014, 8-11 December, London 10

Decision on heterogeneous clouds

Elasticity

Requirements

Elasticity

Relationships

Metrics

monitoring

information

Decision process

Simulate

relationships

effects

Evaluate

control actionsAffected

componentsComponents

to re-evaluate

Application level controlCloud resources

control

UCC 2014, 8-11 December, London 11

Experiments - Settings

M2MDaaS:

STRATEGY CASE avgBufferSize<5 :

minimize(cost)

CONSTRAINT avgBufferSize<50 CONSTRAINT responseTime<50 ms

STRATEGY CASE

responseTime<40ms AND

throughput<20ops/s : scalein

Load relationship

requestsbufferSize

UCC 2014, 8-11 December, London 12

Experiments – Results

Action enforced on local

processing

due to elasticity requirement

Buffer size expected to increase

Compensation action enforced

UCC 2014, 8-11 December, London 13

Experiments – Results

Application cost evolving to accommodate varying load

on two providers

UCC 2014, 8-11 December, London 14

Conclusions and Future Work

Profiting from clouds heterogeneity

– We can fix a variety of issues using the same tools

– Controlling applications as a whole when running over multiple

clouds strengthens the decision capacity

Understanding application distribution

– Understanding relationships which may appear across

distributed components

Ongoing work

– Determining elasticity relationships

– Improving the analysis/decision process

UCC 2014, 8-11 December, London 15

Thank you!

Georgiana Copil

[email protected]://dsg.tuwien.ac.at/staff/ecopil/

Prototypes available http://tuwiendsg.github.io

Distributed Systems GroupVienna University of Technology

Austria

UCC 2014, 8-11 December, London 16

Backup slide – sensitivity analysis