International conference on Multi-cloud computing.

1. Managing Elasticity Accross
Multi-Cloud Providers
1
Fawaz Paraïso, Philippe Merle, Lionel Seinturier
1st International workshop on multi-cloud applications and federated clouds
(2013)
University Lille1 & Inria Lille – Nord Europe (France)

2. 2
Introduction & Motivation
Challenges
Contribution
Implementation
Validation
Conclusion
2
Agenda

3. 3
Introduction & Motivation
Elasticity is the capability to rapidly provision, in some
cases automatically, to quickly scale out, and rapidly
release resources
APP
APP
APP
APP
APP
APP
APP
APP
APP

4. 4
Introduction & Motivation

5. 5
Introduction & Motivation
Cloud provider What happened & why What impactWhen
Amazon [8] The Amazon Elastic
Load Balancing (ELB)
Service down in US-East
region affected the
applications using the
ELB.
21 April 2011 Offline for more
than 10 hours.
Companies
affected:
reddit, Quora, Hoot
Suite.
Windows Azure [9] A networking problem
during a routine software
update interfered with
hosted project
deployment.
13 March 2009 Offline for 22
hours.

6. 6
Introduction & Motivation
Cloud provider Electricity provider
Unavailability: 7.5 hours average per year Unavailability: 15 minutes average per year
Total of 585 hours cost > 71.7 million [1]
[1] International Working Group on Cloud Computing Resiliency. http://www.iwgcr.org

7. 7
Introduction & Motivation
Multiple servers #1
Data centre
Cloud provider
Load balancer

8. 8
Introduction & Motivation
Multiple data centres #2
Load balancer
Data centre A Data centre B Data centre C
Cloud provider

9. 9
Usage of multiple cloud providers in a uniform way.
Multi-Cloud
service
Cloud B Cloud CCloud A
Multiple Clouds #3
Introduction & Motivation

10. 10
Introduction & Motivation
Challenges
Contribution
Implementation
Validation
Conclusion
10
Agenda

11. 11
Challenges
How to guarantee high availability?
How to automate elasticity through multiple clouds?
How to provide transparency?

12. 12
Introduction & Motivation
Challenges
Contribution
Implementation
Validation
Conclusion
12
Agenda

13. 13
Contribution
Manage elasticity problem in Multi-Cloud environment
Provide necessary resources when the system needs
Full instrumentation for monitoring workloads
Unpredictible environment
Why ?What ?

14. 14
Contribution
Overview of the Multi-Cloud-PaaS Architecture

15. 15
Contribution
Deployment of the Multi-Cloud-PaaS Architecture

16. 16
Contribution
Independent of Cloud
Flexible architecture
Automation

17. 17
Introduction & Motivation
Challenges
Contribution
Implementation
Validation
Conclusion
17
Agenda

18. 18
Implementation
Implementation
FraSCAti (SCA Model)
Multi-Cloud-PaaS SCA-Based Components

19. 19
The Multi-Cloud-PaaS (MCP) has been deployed on
ten IaaS/PaaS providers.
Deployment

20. 20
Use case
Application responsible for checking if the JPG format is correct.
App
First case
App
Second case

21. 21
Evaluation
Overhead
Implementation Avg. exec. Time LB overhead
APP 13.93 sec -
APP + LB 14.10 sec 1.45%
There is a negligeable overhead introduced by the LB
First case
Second case
To evaluate the overhead of the LB instance, 10,000 pictures were
sent to the application.
Performance
We generate 134, 021 requests and continuously connects to one
instance of LB
Session rate Concurrency Data rate Failures Avg. Time
850 283 4560 kB/s 0 3 ms
Given the low resources used by the LB, the results obtained in are satisfactory

22. 22
Conclusion
This paper provides solution to manage elasticity
across multiple cloud providers
Independent of Cloud
 Flexible architecture
We plan to
Evaluate the other Multi-Cloud-PaaS architecture
components
Investigate for optimization opportunities of cloud
application deployment

23. 23
Thank you!
Fawaz.paraiso@inria.fr