Stéphanie Challita's PhD Defense Presentation

Inferring Models from Cloud APIs and
Reasoning over Them:
a Tooled and Formal Approach
Presented by:
Stéphanie CHALLITA
PhD committee:
Philippe MERLE Inria Lille – Nord Europe Supervisor
Benoit COMBEMALE University of Toulouse & Inria Rennes Reviewer
Christian PEREZ Inria Lyon Reviewer
Hélène COULLON IMT Atlantique Examiner
Laetitia JOURDAN University of Lille Examiner
Faiez ZALILA Inria Lille – Nord Europe Invited
PhD Defense
December 21, 2018

Stéphanie CHALLITA – PhD Defense 2/109
Cloud computing
Created by Sam Johnston, downloaded from https://en.wikipedia.org/wiki/Cloud_computing

Multi-Cloud computing

4/109Stéphanie CHALLITA – PhD Defense
Problem statement
Deployment
Model
Management
Interface
Service
Model
Public SOAP & REST
IaaS,
PaaS,
SaaS
Public REST
IaaS,
PaaS,
SaaS
Public REST IaaS
Private REST IaaS

Problem statement
Developer
Natural language
is ambiguous!

Problem statement

7/109
To semantically reason over
extracted models of
cloud APIs
Stéphanie CHALLITA – PhD Defense
Thesis objective

1. State of the Art
2. Foundation
 OCCIware
3. Contributions
 Inferring models from cloud APIs: GCP model
 Reasoning on cloud APIs: fclouds framework
4. Conclusion
 Summary
 Perspectives
Outline

Approaches for multi-clouds - Actors
State of the Art Foundation Contributions Conclusion
Cloud
provider
Cloud
developer
Cloud
architect
Use Use
Offer

Programming Space
Provider Space Cloud
provider
Cloud
developer
Cloud
architect
Modeling Space
Semantic Space
Approaches for multi-clouds

AWS
API
OCCI
API
DigitalOcean
API
Provider
Space
Private
Cloud
provider
Public
GCP
API
…
OCCI CIMI …
Public Public

AWS
API
OCCI
API
DigitalOcean
API
Provider
Space
Public Private
Cloud
provider
Public
GCP
API
…
Public
Cloud
Brokers
OCCI CIMI …

AWS
API
OCCI
API
DigitalOcean
API
DigitalOcean
SDK
Programming
Space
GCP
SDK
AWS
SDK
Provider
Space
Public Private
Cloud
provider
Public
GCP
API
Cloud
developerOCCI
SDK
…
…
Public
Cloud
Brokers
OCCI CIMI …

AWS
API
OCCI
API
DigitalOcean
API
DigitalOcean
SDK
Multi-cloud
Libraries
Programming
Space
GCP
SDK
AWS
SDK
Provider
Space
Public Private
Cloud
provider
Public
GCP
API
Cloud
developerOCCI
SDK
…
…
Public
Cloud
Brokers
OCCI CIMI …

Cloud
Metamodel
Cloud
Model
conforms to
represented by
defines
Cloud
Meta-metamodel
conforms to
M0
M1
M2
M3
15/109
Cloud
architect
Model
Code
generation
Static
analysis
Documentation
Transformation

Modeling
Space
Cloud
architect
CloudML
AWS
API
OCCI
API
DigitalOcean
API
DigitalOcean
SDK
Multi-cloud
Libraries
Programming
Space
GCP
SDK
AWS
SDK
Provider
Space
Public Private
Cloud
provider
Public
GCP
API
Cloud
developerOCCI
SDK
…
…
Public
Cloud
Brokers
CAMEL TOSCAOpenTOSCASALOON StratusML
OCCI CIMI …

Issue 1:
Fixed metamodels, not extensible to
support additional concepts

RQ#1: Is it possible to have a solution that allows to
represent all kinds of cloud resources despite their
heterogeneity, and a complete framework for
managing them?
- How to design the cloud developer needs at a high-level
of abstraction?
- How to verify the cloud structural and behavioral
properties before any concrete deployments?
- How to deploy and manage cloud configurations?
Research questions
OCCIware
Research topics: Model-Driven Engineering (MDE), Models@run.time

Issue 2:
Fuzziness of the concepts of the cloud
modeling languages

RQ#2: Is it possible to automatically extract precise
models from cloud APIs and to synchronize them with
the cloud evolution?
- How to provide an accurate description for a cloud API?
- How to correct the existing drawbacks in a cloud API
documentation?
- How to analyze a cloud API documentation?
GCP model
Research topics: API mining, reverse-engineering, NLP
Research questions

Issue 3:
Little attention paid to the semantics

RQ#3: Is it possible to reason on cloud APIs and identify
their similarities and differences?
- How to better understand cloud solutions?
- How to make sure that a cloud solution reflects the
desired behaviour?
- How to ensure an accurate migration from a cloud
solution to another?
fclouds
Research topics: model verification & validation, semantic alignment
Research questions

Thesis vision
Model-Driven Approach
for the Cloud
Formal Approach
for the Cloud
OCCIware
fclouds
Infer
Reason
OCCIGCPAWS

Thesis vision
for the Cloud
OCCIware

 Community-based effort hosted by
 Resource-oriented model and RESTful API
 Everything as a Service, i.e., XaaS
OCCI

OCCI core model
Category
scheme: URI
term: String
title: String [0..1]
Kind Mixin
Action
Entity
id: URI
Resource Link
Attribute
name: String
type: String [0..1]
mutable: Boolean [0..1]
required: Boolean [0..1]
default: String [0..1]
description: String [0..1]
0..1 *
actions 1*
actions
*
mixins
*
entities
1
kind
*
entities
1 target
1 source
*
links
0..1
parent
*
*
depends
*
1 *
attributes
* applies
Source: R.Nyrén, A.Edmonds, A.Papaspyrou, T.Metsch and B.Parák, “Open Cloud Computing Interface-Core,” Open Grid
Forum, In Specification Document GFD.221, Feb. 2016.

OCCIware metamodel

OCCIware tool chain

 OCCIware is a factory to build cloud domain-specific
modeling frameworks
OCCIware use cases

modeling frameworks
OCCIware use cases

Thesis vision
for the Cloud
OCCIware

Thesis vision
for the Cloud
OCCIware
Infer
OCCIGCPAWS

conformsto
Cloud
documentation
Cloud API documentation
Cloud
developer/architect
Cloud
provider
An agreement with the developer on exactly how the system will operate
Cloud documentations are written in natural language
 human errors and/or semantic confusions

Global Vision
 Inferring models from cloud APIs
 Work of API mining, reverse-engineering
HTML Model
 Model refinement (NLP techniques, graphical output…)

Google Cloud Platform (GCP) use case
Is partner withIs adopted by

 Informal heterogeneous documentation
 Imprecise types
 Implicit attribute metadata
 Hidden links
 Redundancy
 Lack of visual support
List of GCP documentation drawbacks

Informal heterogeneous documentation
Available at
https://cloud.google.com/compute/docs/reference/latest/networks
Available at
https://cloud.google.com/dataproc/docs/reference/rest/v1/projects.regions.clusters

Imprecise types

GCP snapshot
 GCP engineers could
update/correct GCP
documentation
 Continuously following up with
GCP documentation is costly
 Snapshot of GCP API
A
Snapshot
GCP
HTML pages
GCP
documentation

GCP crawler & GCP model
GCP
Crawler
A B
Snapshot
GCP
HTML pages
GCP
documentation
 GCP Crawler to extract all GCP resources, their
attributes and actions
 GCP Model for a better description of the GCP
resources
GCP
Model
C

GCP
Crawler GCP
Model
A B
Snapshot
GCP
HTML pages
GCP
documentation
C
 GCP Crawler to extract all GCP resources, their
attributes and actions
 GCP Model for a better description of the GCP
resources and for reasoning over them
No more Informal Heterogeneous Documentation
OCCIware
Metamodel
GCP
configuration
conforms to
represented by
Ecore
Metamodel
conforms to
M0
M1
M2
M3
GCP
model
GCP
doc
conforms to

Model transformations
Implicit Attribute
Metadata Detection
Link Identification
Redundancy Removal
Model
Transformations
Type Refinement
Model Visualization
GCP
Crawler
A B
Snapshot
GCP
HTML pages
GCP
documentation
GCP
Model
C

58/109
 By adopting the data type system proposed by
OCCIware metamodel
 defining regular expressions
 using the EMF validator to check the type constraints that are
attached to the attributes
Type refinement

59/109
 By adopting the data type system proposed by
OCCIware metamodel
 defining regular expressions
 using the EMF validator to check the type constraints that are
attached to the attributes
Type refinement
 If the type of an attribute in the documentation is
string and the description explains that this is an
email address, we apply the email validation
constraint:
 STRINGTYPE + this regular expression:
^[A-Z0-9._%+-]+@[A-Z0-9.-]+.[A-Z]{2,6}$
No more Imprecise Types

60/109
 To explicitly store information into additional attributes
defined in the ATTRIBUTE concept of our GCP MODEL
 We use Natural Language Processing (NLP) techniques
Word Tagging/Part-of- Speech (PoS)
 We declare pre-defined tags for some GCP specific
attribute properties:
 mu tab le = tru e if [In p u t -O n ly ]
 mu tab le = false if [O u tp u t - on ly ]/ read on ly
 required = true if [Required]
 req u ired = false if [O ption al ]
 d efau lt = X if Th e d efau lt valu e is X
Implicit attribute metadata detection

Model visualization
No more Lack of Visual Support Built via OCCIware
designer

66/109
 Documentation spread over many pages
– Recursive parsing one must deeply explore the documentation to completely
define all the required concepts
Challenges

67/109
 Huge and tough documentation analysis
– Finely observe the structure of the page to correctly design the crawler
– Read the descriptions to carefully design the rules
Challenges

68/109
 Huge and tough documentation analysis
– Finely observe the structure of the page to correctly design the crawler
– Read the descriptions to carefully design the rules
 Huge and tough model analysis
– Verify the automatically set values to incrementally refine the knowledge
extraction
Challenges

Analysis
 How many resources are provided by GCP
documentation?
 Why GCP documentation is heterogeneous?
 To what extend GCP resources have redundant
characteristics?
 How GCP model allows us to factorize them?

How many resources are provided by GCP
documentation?

Why GCP documentation is heterogeneous?
Runtime Config
Cloud User
Account &
GCP
Model

To what extend GCP resources have
redundant characteristics?

How GCP model allows us to factorize
them?

Thesis vision
for the Cloud
OCCIware
Infer
OCCIGCPAWS

Thesis vision
for the Cloud
Formal Approach
for the Cloud
OCCIware
fclouds
Infer
Reason
OCCIGCPAWS

Need to reason on the common principles that
cloud solutions must adhere to
Exploring the semantic space
Multi-cloud Libraries
Modeling
Space
Programming
Space
Provider
Space
Cloud Brokers Cloud
provider
Cloud
developer
Cloud
architect
Model-Driven Approaches for the Cloud (MDAC)
Semantic
Space
Formal Approaches for the Cloud

Need to reason on the common principles that
cloud solutions must adhere to
Exploring the semantic space
[1] K. Yongsiriwit, M. Sellami, and W. Gaaloul, “A Semantic Framework Supporting Cloud Resource Descriptions Interoperability,” in 2016 IEEE
9th International Conference on Cloud Computing (CLOUD). IEEE, 2016, pp. 585–592.
[2] N. Loutas, E. Kamateri, and K. Tarabanis, “A Semantic Interoperability Framework for Cloud Platform as a Service,” in 2011 IEEE Third
International Conference on Cloud Computing Technology and Science (CloudCom). IEEE, 2011, pp. 280–287.
Multi-cloud Libraries
Modeling
Space
Programming
Space
Provider
Space
Cloud Brokers Cloud
provider
Cloud
developer
Cloud
architect
Model-Driven Approaches for the Cloud (MDAC)
Semantic
Space
fcloudsPSIF [2][1]

Need for the semantic space
Mathematical
specification
Formal
Validation
Reasoning
One
interpretation
Accuracy
Earlier error detection,
cheaper correction

fclouds framework
fclouds is a framework for providing formal
specifications of cloud APIs & reasoning over them
 Catalog of cloud formal models
 Based on a formal language

Formalization of OCCI core concepts & CRUD operations
in Alloy
fclouds formal language

82/109
 Formalization of OCCI core concepts in Alloy
fclouds static semantics
Extension Configuration
Resource Link
KindAction
DataType
Attribute
Time
source
target
links
use
kinds
types
resources
action

83/109
 Concepts are modeled as signatures
Resource Link
KindAction
DataType
Attribute
Time
source
target
links
use
kinds
types
resources
action
sig Configuration {
use : set Extension ,
resources : set Resource -> Time
}

84/109
 Concepts are modeled as signatures
 Time concept is added to distinguish between
mutable and immutable fields
Resource Link
KindAction
DataType
Attribute
Time
source
target
links
use
kinds
types
resources
action
sig Configuration {
use : set Extension ,
resources : set Resource -> Time
}

 Formalization of OCCI behavioral specification1 in Alloy
 Operations are modeled as predicates
OCCIAPI
Create
Retrieve
Update
Delete
1. R. Nyrén, A. Edmonds, T. Metsch and B. Parák, “Open Cloud Computing Interface - HTTP Protocol,” Open Grid Forum,
In Specification Document GFD.223, Feb. 2016.
fclouds operational semantics

 Formalization of OCCI behavioral specification1 in Alloy
 Operations are modeled as predicates
 Time concept is added to distinguish between pre-
states and post-states
OCCIAPI
Create
Retrieve
Update
Delete
1. R. Nyrén, A. Edmonds, T. Metsch and B. Paràk, “Open Cloud Computing Interface - HTTP Protocol,” Open Grid Forum,
In Specification Document GFD.223, Feb. 2016.

pred CreateResource [ config : Configuration, resourceId : String, kind :
Kind , t, t ’ : Time ] {
/ / preconditions at instant t
no resource : config.resources.t |
resource.id = resourceId
kind in config.use.kinds
}

pred CreateResource [ config : Configuration, resourceId : String, kind :
Kind , t, t ’ : Time ] {
/ / preconditions at instant t
no resource : config.resources.t |
kind in config.use.kinds
/ / postconditions at instant t ’
one resource : Resource {
resource . kind = kind
config.resources. t ’ = config.resources.t + resource }
}

fclouds properties
 Consistency
 Sequentiality
 Reversibility
 Conformance to HTTP 2 protocol
 Idempotence
 Safety

fclouds properties - sequentiality
Definition:
“Two cloud API operations are sequential when one cannot
happen if the other one did not happen at the time before ”

Definition:
“Two cloud API operations are sequential when one cannot
happen if the other one did not happen at the time before ”
No counterexample
t1 t2
Update-VM
t0
Cores: 2
Memory: 2 GB
Disk: 256 GB
Cores: 4
Memory: 2 GB
Disk: 256 GB
assert RetrieveResourceThenUpdateResource {
... }

Pairs of sequential OCCI operations
Create Retrieve Update Delete
Create
Retrieve
Update
Delete

Catalog of cloud formal specifications
IaaS
PaaS
IoT
Transverse cloud
concerns
MoDMaCAO [CLOSER 2018]
OMCRI [IoT 2018]
CoT [CoopIS 2017]
Platform [OGF 2016]
Infrastructure [OGF 2016]
CRTP [OGF 2016]
SLA [OGF 2016]
Monitoring [OGF 2016]
Cloud Simulation [EDGE 2017]
Cloud Elasticity [CLOUD 2017]
[OCCIware Deliverable 2.4.1]
[IC2E 2018]
[CLOUD 2016]

 Template-based approach
 Alloy concept = templates for each OCCI concept
conforms to
Alloy generator
OCCI
extension
Alloy
specification
fclouds
specification

Alloy generator

97/109
 Verification of fclouds properties via Alloy Analyzer
Verification of properties

98/109
 Verification of fclouds properties via Alloy Analyzer
 Definition & validation of domain-specific properties
Example
 In OCCI Infrastructure:
Compute NetworkNetworkInterface
assert NetworkInterfaceBetweenComputeAndNetwork {
... }
Verification of properties

pred ComputeMapInstance [ c : one Compute, i : one Instance ] {
i.name = c.occicomputehostname
i.machinetype.isSharedCpu = c.occicomputeshare
i.machinetype.memoryMb = mul [ 1024, c.occicomputememory ]
…
}
99/109
 Ensure semantic alignment
Example
 An instance at GCP is a compute at OCCI
Formal transformation rules
Compute
GCP
configuration
OCCI
configuration
resources
resources
String
Integer
Boolean
Instance

Thesis vision
for the Cloud
Formal Approach
for the Cloud
OCCIware
fclouds
Infer
Reason
OCCIGCPAWS

Global scenario
GCP
Model
GCP.als
Infer
Reason
transformation
Model-driven engineering
Integrated formal methods

 RQ#1: Is it possible to have a solution that allows to represent all
kinds of cloud resources despite their heterogeneity, and a complete
framework for managing them?
A model-driven framework to deal with all kinds of cloud
resources with OCCIware
 RQ#2: Is it possible to automatically extract precise models from
cloud APIs and to synchronize them with the cloud evolution?
Inferring precise cloud models from online documentations
and the GCP use case
 RQ#3: Is it possible to reason on cloud APIs and identify their
similarities and differences?
Reasoning on cloud models with fclouds
Summary

 Following the evolution of GCP API
Perspectives

 Developing more semantic alignment between cloud
APIs
Perspectives

 Developing more semantic alignment between cloud
APIs
 Improving the management of cloud applications
with OCCI and TOSCA
Perspectives

 Extending the properties of formal cloud APIs
Perspectives

 Extending the properties of formal cloud APIs
 Exploring the use of formal models to address the
challenges of other distributed systems (IoT, edge
computing, etc.)
Perspectives

Stéphanie CHALLITA
stephanie.challita@inria.fr
researchers.lille.inria.fr/schallit
Thank you!
Questions?

International journal
Faiez Zalila, Stéphanie Challita, Philippe Merle. “Model-Driven Cloud Resource Management with OCCIware.” Future
Generation Computer Systems (FGCS), 2018. (under review)
International conferences
Stéphanie Challita, Faiez Zalila, Philippe Merle. “Specifying Semantic Interoperability between Heterogeneous Cloud
Resources with the fclouds Formal Language”. IEEE International Conference on Cloud Computing (CLOUD). 2018.
Stéphanie Challita, Faiez Zalila, Christophe Gourdin, Philippe Merle. “A Precise Model for Google Cloud Platform" .
IEEE International Conference on Cloud Engineering (IC2E). 2018.
Fabian Korte, Stéphanie Challita, Faiez Zalila, Philippe Merle, Jens Grabowski. “Model-Driven Configuration
Management of Cloud Applications with OCCI”. International Conference on Cloud Computing and Services Science
(CLOSER). 2018.
Faiez Zalila, Stéphanie Challita, Philippe Merle. “A Model-Driven Tool Chain for OCCI”. International Conference on
Cooperative Information Systems (CoopIS). 2017.
Stéphanie Challita, Fawaz Paraiso, Philippe Merle. “Towards Formal-based Semantic Interoperability in Multi-Clouds:
The fclouds Framework’’. IEEE International Conference on Cloud Computing (CLOUD). 2017.
Stéphanie Challita, Fawaz Paraiso, Philippe Merle. “A Study of Virtual Machine Placement Optimization in Data
Centers”. International Conference on Cloud Computing and Services Science (CLOSER). 2017.
Fawaz Paraiso, Stéphanie Challita, Yahya Al-dhuraibi, Philippe Merle. “Model-Driven Management of Docker
Containers”. IEEE International Conference on Cloud Computing (CLOUD). 2016.
Award
Awardee of L’Oréal – UNESCO For Women In Science program

Stéphanie Challita's PhD Defense Presentation

Recommended

Recommended

More Related Content

What's hot

What's hot (17)

Similar to Stéphanie Challita's PhD Defense Presentation

Similar to Stéphanie Challita's PhD Defense Presentation (20)

Recently uploaded

Recently uploaded (20)

Stéphanie Challita's PhD Defense Presentation