Transcript Slide 1

Mastering Intelligent Clouds
Engineering Intelligent Data Processing
Services in the Cloud
Sergiy Nikitin,
Industrial Ontologies Group,
University of Jyväskylä, Finland
Presented at
ICINCO 2010 conference
Funchal, Madeira
Contents
• Background on Cloud Computing
• Extending cloud computing stack
• UBIWARE platform
• Data Mining services in the Cloud
• Conclusions
Cloud Computing: already on the market
•
•
•
•
•
•
•
•
•
SalesForce.com (SFDC)
NetSuite
Oracle
IBM
Microsoft
Amazon EC2
Google
etc.
(for a complete survey see Rimal et al., 2009)
Cloud Computing stack
Cloud
computing
stack
Application as a Service
SaaS
Application
(business logic)
Services (Payment,
Identity, Search)
PaaS
Solution stack (Java,
PHP, Python, .NET)
Structured storage
(e.g. databases)
Raw data storage and
network
OS-virtualization
IaaS
Virtualization Machine
Hardware configuration
What add-value can we
offer to the PaaS level?
Autonomic Computing
• A vision introduced by IBM in 2003 (Kephart et al.)


software components get a certain degree of selfawareness
self-manageable components, able to “run themselves”
• Why?


To decrease the overall complexity of large systems
To avoid a “nightmare of ubiquitous computing” – an
unprecedented level of complexity of information
systems due to:
• drastic growth of data volumes in information systems
• heterogeneity of ubiquitous components, standards, data
formats, etc.
Intelligence as a Service in the cloud
Services (Payment,
Identity, Search)
PaaS
Solution stack (Java,
PHP, Python, .NET)
Structured storage
(e.g. databases)
Agent-driven service API
Configuration management
Data adaptation
Intelligent services
Solution stack
Domain models
UBIWARE
• Smoothly integrate with the infrastructure
• Build stack-independent solutions
• Automate reconfiguration of the solutions
UBIWARE platform
UBIWARE Agent
Beliefs storage
Pool of Atomic
Behaviours
.class
S-APL
RAB
RAB
RAB
Blackboard
Data
RAB
Role Script
S-APL
repository
API extension: OS perspective
Cloud Platform Provider
Virtual machine
PCA
Customer
applications
and services
Extended API
SW Platform
PCA – Personal Customer Agent
PMA – Platform Management Agent
PMA
Data Adaptation as a Service
Cloud Platform Provider
Virtual machine
PMA
SW Platform
Customer
applications
and services
Extended API
PCA
Data Service
Files
Adapter
Agent
PCA – Personal Customer Agent
PMA – Platform Management Agent
DB/KB
Platform-driven service execution in the cloud
Cloud Platform Provider
Virtual machine
Virtual machine
Service
execution
environment
Customer
applications
API
SW Platform
PCA
API
PCA – Personal Customer Agent
PMA – Platform Management Agent
PMA
Agent-driven PaaS API extension
Agent-driven flexible intelligent service API
Agent-driven Adapters
Smart data source connectivity
Configurable data transformation
Agent-driven intelligent services
Configurable model
Service mobility
Proactive self-management
Smart cloud stack
Stack control and updates
Failure-prone maintenance
Embedded and remote services
Smart Ontology
Domain models
Standards & compatibility
System configuration and policies
User applications in cloud
Proactive adapter management
Intelligent services: PaaS API extension
Agent-driven intelligent services
Configurable model
Service mobility
Proactive self-management
User applications in cloud
Agent-driven flexible intelligent service API
Agent-driven data mining services
 Data mining applications are capabilities

Agents can wrap them as services
 PMML language - a standard for DM-model representations
 Data Mining Group. PMML version 4.0. URL
http://www.dmg.org/pmml-v4-0.html
Agent service
Input
Vector
Model
DM model
Output
DM result
PMML*: data mining model descriptions
PMML model
Header
Version and timestamp
Model development environment information
Data dictionary
Definition of: variable types, valid,
invalid and missing values
Data Transformations
Normalization, mapping and discretization
Data aggregation and function calls
Model
Description and model specific attributes
Mining schema
Definition of: usage type, outlier and
missing value treatment and replacement
Targets
Score post-processing - scaling
Definition of model architecture/parameters
PMML* - Predictive Model Markup Language (www.dmg.org/pmml-v3-0.html)
Data mining service types
Input
Fixed model service
Model
Output
Vector to be classified:
alarm message:
V1={0.785, High, node_23}
Paper machine alarms
classifier neural
network model (M1)
Vector class of V1 is:
“Urgent Alarm”
according to model M1
Inputs
Model
Outputs
Model player
Model M1 assigned
Paper machine alarms
classifier neural
network model (M1)
Vector class of V1 is:
“Urgent Alarm”
according to model M1
Model
Output
Model constructor
Model M1 parameters
Set up a model
M1
Model player service
Vector to be classified:
alarm message:
V1={0.785, High, node_23}
Input
Model construction service
Learning samples and the
desired model settings
A use case for data mining service stack
 A “Web of Intelligence” case:
Input
1
2
3
4
Model
Output
Pattern of learning data
to be collected:
?V={?p1, ?p2, ?p3}
Distributed query
planning and
execution
A set of learning
samples (vectors)
Learning samples and the
desired model settings
Model constructor
Model M1 parameters
Model player
Model M1 assigned
Paper machine
alarms classifier
neural network
model (M1)
Vector class of V1 is:
“Urgent Alarm”
according to model M1
Set up a model
M1
Vector to be classified:
alarm message:
V1={0.785, High, node_23}
Data Mining services in UBIWARE
Ontology construction
Data Mining
service
Model construction
service
Mining method
Computational
service
Fixed model
service
Supervised
Learning
Model player
service
Neural networks
Industry
Process Industry
Paper industry
Electrical
Engineering
Power networks
Problem domain
Power plant
Unsupervised
learning
Clustering
kNN
UBIWARE in cloud computing stack
Cloud computing
stack
Example application
UBIWARE for control and management in cloud
Semantic Business
Scenarios
Applications
and Software
as a Service
Application as a
Service
DM model wrapped as a
service for paper industry
Application
(business logic)
DM model for paper
industry
Services (Payment,
Identity, Search)
Platform
as a
service
Solution stack (Java,
PHP, Python, .NET)
Structured storage
(e.g. databases)
Data Mining
service player
Agent-driven
service API
Domain-specific
components as services
Domain model (Ontology) &
components
Cross-domain
Middleware
components
Connectors,
Adapters
RABs, Scripts
Componentization &
Servicing
Cross-layer
configuration &
management
mechanisms
Raw data storage
and network
OS-virtualization
Infrastructure
as a service
Virtualization
Machine
Hardware
configuration
Technologies in cloud
Conclusions
• Web intelligence as a cloud service
• Ubiware is a cross-cutting management and
configuration glue
• Advanced data adaptation mechanisms as cloud
services


A competitive advantage for cloud providers
Seamless data integration for service consumption and
provisioning
• Autonomous agents as a Service (A4S)

Supply any resource with the “autonomous manager”
Thank you!