Presented at Funchal, Portugal by Frederico Alvares (EasyVirt). Full paper available from https://hal.archives-ouvertes.fr/hal-01705248

1. A Model-based Architecture for
Autonomic and Heterogeneous Cloud
Systems
Frederico Alvares
EasyVirt, Nantes
Zakarea Al Shara
Berger-Levrault, Montpellier
Jonathan Lejeune
Université Sorbonne (Paris 6)
Hugo Bruneliere, Thomas Ledoux
IMT Atlantique & LS2N (CNRS), Nantes

2. Context and Motivation
Challenges to efficiently design, (re)configure and monitor Cloud systems
● Cloud systems are heterogeneous
○ Many resources of varied natures (software and/or physical)
○ Need for support in a similar way resources coming from different layers
● Cloud systems are highly dynamic
○ Clients can book/release “elastic” virtual resources at any moment
○ Need for transparent reflection & support of this dynamicity/elasticity
● Cloud systems are becoming complex and cannot be handled manually
○ Configuration and monitoring, but also runtime behavior to guarantee SLA contracts
○ Need for decision-making and reconfiguration support
2

3. Motivating Example
● End-users: students, faculty
members, admin. staff, etc.
● SaaS: keep low response times
● IaaS: keep resources available
● EaaS: promote green energy usage
● Dynamical environment
○ On-demand provisioning makes Cloud
susceptible to short-term variations
○ e.g., overload situations during practical
sessions
SaaS
Energy as a Service
Cloud Client
E-Learning
BrownGreen
IaaS
Large VMSmall VM
SLA:
response time
SLA:
availability
SLA:
% green energy
3

4. Generalizing, the goal is to...
● Autonomically reconfigure the system to have the best balance between
○ Costs, which are related to services consumed/bought from providers
○ Revenues, which refer to services offered/sold to clients
● While complying with SLA
Cloud Client
SaaS
IaaS
Energy as a Service
consume produce
consume produce
consume produce
SLA
SLA
SLA
produceconsume
SLA
SLA
XaaS
Layer n+1
Layer n-1
4

5. CoMe4ACloud
Constraints and Model Engineering for Autonomic Clouds
A generic and extensible solution for the autonomic management of Cloud
services, independently from the Cloud layer(s)
● A constraint-based model has been proposed [Best Paper Award, CLOSER’2017]
○ But it lacks proper model-based architecture and a reusable modeling language
Contributions
● A model-based architecture for generic autonomic XaaS management
● A related XaaS core modeling language, possibly supporting any of the
possible Cloud layers
● An Eclipse-based tooling support
5

6. Outline
1. Architecture Overview
2. Core Modeling Language
3. Tooling Support and Implementation
4. Related Work
5. Final Remarks
6

7. Architecture Overview
7

8. Architecture Overview
Topology
Metamodel
conforms to
Constraint
Program (SLA)
Configuration
Metamodel
Topology
Model t
conforms to
Configuration
Model c0
refers to
generation
Configuration
Model c1
refers to
conforms to
Cloud
Expert
Cloud
Administrator
Runtime
Design time
Cloud
System s
represents
represents
<<TOSCA>>
Topo./Config. Model tcX’
TOSCA Tool
CoMe4ACloud
External Solution(s)
transformation
transformation
8
, cX

9. Core Modeling Language
9

10. Topology Metamodel
Topology: ELearning-IaaS
node_types:
InternalComponent:
PM:
derived_from: InternalComponent
properties:
...
VM:
derived_from: InternalComponent
...
Cluster:
derived_from: InternalComponent
properties:
constant ClusterConsOneCPU:
type: integer
constant ClusterConsOneRAM:
type: integer
constant ClusterConsMinOnePM:
type: integer
variable ClusterNbCPUActive:
type: integer
equal: Sum(Pred, PM.PmNbCPUAllocated)
variable ClusterCurConsumption:
type: integer
equal: ClusterConsMinOnePM * NbLink(Pred)
+ ClusterNbCPUActive * ClusterConsOneCPU
+ ClusterConsOneRAM
* Sum(Pred, PM.PmSizeRAMAllocated)
10

11. Topology Metamodel
...
Power:
derived_from: ServiceProvider
properties:
constant PowerCapacity:
type: integer
variable PowerCurConsumption:
type: integer
equal: Sum(Pred, Cluster. ClusterCurConsumption)
constraints:
PowerCurConsumption less_or_equal: PowerCapacity
relationship_types:
VM_To_PM:
valid_source_types: VM
valid_target_types: PM
PM_To_Cluster:
...
Cluster_To_Power:
...
11

12. Configuration Metamodel
Configuration:
identifier: ElarningSystem_0
topology: ELearning-IaaS
...
Node Power0:
type: ELearning-IaaS.Power
properties:
PowerCapacity: 1500
PowerCurConsumption: 0
...
Node Cluster0:
type: ELearning-IaaS.Cluster
properties:
ClusterCurConsumption: 0
ClusterNbCPUActive: 0
ClusterConsOneCPU: 1
ClusterConsOneRAM: 0
ClusterConsMinOnePM: 5
...
Node PM0:
type: ELearning-IaaS.PM
...
Relationship PM_To_Cluster0:
type: ELearning-IaaS.PM_To_Cluster
constant: true
source: PM0
target: Cluster0
12

13. Tooling Support
13

14. Tooling Support
14

15. Application to the Motivating Example
● Design of the SaaS layer
○ XaaS modeling language
● Interoperability
○ Eclipse Winery
(TOSCA-based GUI)
● Runtime adaptation
○ Constraint-based
decision-making architecture
○ Reconfiguration from current
state c0 to another one c1
15

16. 16

17. Current Status & Related Work
17

18. Current Status
● Runtime (synchronization with the actual systems)
○ Implementation of a common adaptor interface for each target system
○ Currently implemented: Amazon EC2, OpenStack
● Scalability (vs. genericity)
○ Solutions found in < 10s for several hundreds of nodes (in a single model)
○ Alternative - hierarchizing the constraints to avoid combinatorial explosion
● Language V&V
○ The current version comes with support for basic syntactic validation
○ We plan to provide support to verify the correctness of the topologies and/or configuration
■ e.g., detect conflicting constraints
● Integration with Cloud standards
○ Possible proposition of XaaS language’s features to the TOSCA standardization group
■ e.g., support for (runtime) expressions/constraints
18

19. Related Work
Genericity UI/
Language
Interoperability Runtime
Support
APIs/DevOps ✅ ✅
Cloudify ✅ ✅ ✅
Brooklyn ✅ ✅ ~
4CaaSt (Nguyen et al., 2011) ✅
ARTIST-CAML (Bergmayr et al., 2014) ✅ ✅
mOSAIC (Sandru et al., 2012) ✅ ~
Stratus ML (Hamdaqa and Tahvildari, 2015) ✅ ✅
CloudMF (Ferry et al., 2013, 2014) ✅ ~
PaaSage-CAML (Achilleos et al., 2015) ✅ ~
SRL (Domaschka et al., 2014) ~ ✅ ✅
Saloon (Quinton et al., 2016) ✅ ✅
ARCADIA (Gouvas et al., 2016) ✅ ✅ ~
Descartes (Kounev et al., 2016) ✅ ✅
MODAClouds (Pop et al., 2016) ✅ ✅
(Garcia-Galan et al., 2014) ✅ ✅
CoMe4ACloud ✅ ✅ ~ ~ 19

20. Final Remarks
20

21. Conclusions and Perspectives
● Contribution: an architecture for the autonomous runtime management of
heterogeneous Cloud systems
○ Generic XaaS models that can possibly interoperate with standards (e.g., TOSCA)
○ Suitable interface, via a XaaS modeling language, to both Cloud experts and administrators
○ Constraint-based decision-making tool to automatically obtain system configurations
respecting specified SLA contracts at runtime
● Perspectives:
○ Application our approach to other fields of Cloud Computing (e.g., Fog)
○ Machine Learning techniques to guide or assist the constraint specification process
21

22. Obrigado
Thank you!
Frederico Alvares
frederico.alvares@easyvirt.com
22