Supporting Autonomic Management of Clouds: Service-Level-Agreement, Cloud Monitoring and Similarity Learning by Rafael Brundo Uriarte
[email protected] Under the Supervision of: Prof. Rocco De Nicola and Prof. Francesco Tiezzi
Doctoral Thesis Defense - March 30th, 2015 - Lucca, Italy
Contents 1
Introduction
2
SLA for Clouds
3
Cloud Monitoring
4
Similarity Learning
5
Polus Framework
6
Conclusions Rafael Brundo Uriarte
1/51
Introduction
Introduction
Rafael Brundo Uriarte
2/51
Cloud Computing
Introduction
Rafael Brundo Uriarte
3/51
Cloud Characteristics
I
Services
I
Heterogeneity
I
Virtualization
I
Large-Scale
I
Complexity
Introduction
Rafael Brundo Uriarte
4/51
Autonomic Computing
Introduction
Rafael Brundo Uriarte
5/51
Autonomic Computing
Introduction
Rafael Brundo Uriarte
6/51
Knowledge for the Self-Management
I
Policies
I
Service Definition and Objectives
I
Status of the Cloud and Services
I
Specific Knowledge
Introduction
Rafael Brundo Uriarte
7/51
Challenge and Scope
Introduction
Rafael Brundo Uriarte
8/51
Research Questions Research Question 1 How to describe services and their objectives in the cloud domain?
Research Question 2 What is data, information, knowledge and wisdom in the autonomic cloud domain?
Research Question 3 How to collect and transform operational data into useful knowledge without overloading the autonomic cloud? Introduction
Rafael Brundo Uriarte
9/51
Research Questions
Research Question 4 How to produce a robust measure of similarity for services in the domain and how can this knowledge be used?
Research Question 5 How to integrate different sources of knowledge and feed the autonomic managers?
Introduction
Rafael Brundo Uriarte
10/51
SLA for Clouds
SLA for Clouds
Rafael Brundo Uriarte
11/51
Service-Level-Agreement (SLA)
I
Contract
I
Service Description
I
Quality-of-Service
I
Formalism
I
Guarantees!
SLA for Clouds
Rafael Brundo Uriarte
12/51
SLA for Cloud Computing - SLAC
I
Domain Specific
I
Multi-Party
I
Deployment Models
I
Formalism
I
Ease-of-Use
SLA for Clouds
Rafael Brundo Uriarte
13/51
Formal Business General
Yet Another SLA Definition Language? Features
WSOL
WSLA
SLAng
WSA
SLA*
SLAC
Deployment Models
Broker Support
-
-
-
-
-
Pricing Schemes
-
Semantics
-
-
-
-
Verification
-
-
-
-
“” feature covered “” feature partially covered “-” no support
SLA for Clouds
Rafael Brundo Uriarte
14/51
Main Concepts
I
Predefined Metrics - Involved Parties and Unit
I
Intervals for Metrics - Template and Variations
I
Groups - Multiple Service, Community Cloud
I
Constraint Solving Problem
SLA for Clouds
Rafael Brundo Uriarte
15/51
Example
SLA for Clouds
Rafael Brundo Uriarte
16/51
Business Aspects
I
Business Actions
I
Flat and Variable Models
I
Pricing Schemes - Exchange, Auction, Tender, Bilateral, Fixed, Posted
SLA for Clouds
Rafael Brundo Uriarte
17/51
Implementation
I
Editor for SLAs (Ecplise-based using Xtext)
I
SLA Evaluator (Z3 Solver)
I
Integration with the Monitoring System
SLA for Clouds
Rafael Brundo Uriarte
18/51
Cloud Monitoring
Cloud Monitoring
Rafael Brundo Uriarte
19/51
DIKW in the Domain
I
Data
I
Information
I
Knowledge
I
Wisdom
Cloud Monitoring
Rafael Brundo Uriarte
20/51
Cloud Monitoring
The Role of the Monitoring System in Clouds: I
Collect data and Provide Information and Knowledge
I
No Wisdom - Related to Decision-Making
I
Sensor of MAPE-K Loop
Cloud Monitoring
Rafael Brundo Uriarte
21/51
Related Works Property
PCMONS
Monalytics
Lattice
Wang
Cloud
-
-
-
Autonomic Integration
-
-
-
-
Scalability
-
Adaptability
-
-
Resilience
-
-
-
-
Timeliness
-
-
Extensibility
“” feature covered “” feature partially covered “-” no support Cloud Monitoring
Rafael Brundo Uriarte
22/51
Panoptes
I
Multi-agent system
I
Monitoring in different levels
I
Monitoring Modules - What needs to be monitored and how to process the data
Cloud Monitoring
Rafael Brundo Uriarte
23/51
Architecture
Cloud Monitoring
Rafael Brundo Uriarte
24/51
Architecture
Communication:
Adaptativeness:
I
Publish/Subscribe
I
Priority for Modules
I
Private Message
I
Change of Roles
Cloud Monitoring
Rafael Brundo Uriarte
25/51
Architecture: Autonomic Integration
I
Urgency Mechanism
I
Decentralised Architecture
I
On-the-Fly Configuration
I
Multiple Abstractions
Cloud Monitoring
Rafael Brundo Uriarte
26/51
Experiments
I
Self-Protection System
I
Urgency Mechanism
I
Scalability
Cloud Monitoring
Rafael Brundo Uriarte
27/51
Similarity Learning
Similarity Learning
Rafael Brundo Uriarte
28/51
Specific Knowledge
Generation of Knowledge for a Specific Purpose, i.e. not applicable in all clouds. For example, similarity. But what is similarity? I
How much an object (service) resembles other
Similarity Learning
Rafael Brundo Uriarte
29/51
Applications of Similarity
Applications in the Domain: Cluster Services: I
I
I
Anomalous Behaviour Detections
I
Service Scheduling
I
Application Profiling
I
SLA Risk Assessment
Group Similar Services Different Algorithms (K-Means, PAM, EM)
Similarity Learning
Rafael Brundo Uriarte
30/51
Domain Requirements
I
Categorical Characteristics of Services
I
On-line Prediction
I
Large Number of Characteristics
I
Fast Prediction
Similarity Learning
Rafael Brundo Uriarte
31/51
Random Forest
Clustering with Random Forest I
Originally Developed for Classification
I
Calculate the Similarity
I
Clustering Algorithm (PAM)
Similarity Learning
Rafael Brundo Uriarte
32/51
Similarity Using RF: Criteria
Similarity Learning
Rafael Brundo Uriarte
33/51
Problems
I
Similarity Matrix (Big Memory Footprint)
I
Re-cluster on Every New Observation
I
Cannot be Used in the Domain
Similarity Learning
Rafael Brundo Uriarte
34/51
Solution: RF+PAM
Similarity Learning
Rafael Brundo Uriarte
35/51
Solution: RF+PAM
Similarity Learning
Rafael Brundo Uriarte
36/51
Experiments I
Compared the performance of our algorithm to other 2 methodologies
I
Compared the performance of RF+PAM with the standard off-line similarity learning
I
Use Case: I Scheduler deploys together the most dissimilar services I Similarity based on their SLAs
Similarity Learning
Rafael Brundo Uriarte
37/51
Polus Framework
Polus Framework
Rafael Brundo Uriarte
38/51
Polus Framework
Polus Framework
Rafael Brundo Uriarte
39/51
Use Case
Polus Framework
Rafael Brundo Uriarte
40/51
Use Case
Polus Framework
Rafael Brundo Uriarte
41/51
Use Case
Polus Framework
Rafael Brundo Uriarte
42/51
Conclusions
Conclusions
Rafael Brundo Uriarte
43/51
Summary
Conclusions
Rafael Brundo Uriarte
44/51
Research Questions Research Question 1 How to describe services and their objectives in the cloud domain? SLAC Research Question 2 What is data, information, knowledge and wisdom in the autonomic cloud domain? DIKW Hierarchy Research Question 3 How to collect and transform operational data into useful knowledge without overloading the autonomic cloud? Panoptes Conclusions
Rafael Brundo Uriarte
45/51
Research Questions
Research Question 4 How to produce a robust measure of similarity for services in the domain and how can this knowledge be used? RF+PAM
Research Question 5 How to integrate different sources of knowledge and feed the autonomic managers? Polus Framework
Conclusions
Rafael Brundo Uriarte
46/51
Limitations
I
Intelligence of Autonomic Managers
I
Wide Range of Specific Knowledge
I
Off-line Training of RF+PAM
Conclusions
Rafael Brundo Uriarte
47/51
Contributions
I
A theoretical and practical framework for the generation and provision of knowledge for the autonomic management of clouds (Polus Framework): I I I
Conclusions
SLAC - SLA Definition and Evaluation Panoptes - Monitoring RF+PAM - Similarity Learning
Rafael Brundo Uriarte
48/51
Publications 1.
R. B. Uriarte, S. Tsaftaris and F. Tiezzi. Service Clustering for Autonomic Clouds Using Random Forest. In Proc. of the 15th IEEE/ACM CCGrid [In Press], 2015.
2.
R.B. Uriarte, F. Tiezzi, R. De Nicola, SLAC: A Formal Service-Level-Agreement Language for Cloud Computing. In IEEE/ACM 7th International Conference on Utility and Cloud Computing (UCC), 2014.
3.
R.B. Uriarte, C.B. Westphall, Panoptes: A monitoring architecture and framework for supporting autonomic Clouds, In Proc. of the 16th IEEE/IFIP Network Operations and Management Symposium (NOMS), 2014.
4.
R.B. Uriarte, S.A. Chaves, C.B. Westphall, Towards an Architecture for Monitoring Private Clouds. In IEEE Communications Magazine, 49, pages 130-137, 2011.
Conclusions
Rafael Brundo Uriarte
49/51
Future Works
I
Dynamic SLAs
I
Negotiation of SLAs
I
Cloud Case Study
Conclusions
Rafael Brundo Uriarte
50/51
Thank you! Questions?
Rafael Brundo Uriarte
[email protected] Conclusions
Rafael Brundo Uriarte
51/51
SLAC - Expressivity
I
Core Language
I
Extensions - Business Aspects
I
Formal Definition for Extensions
Conclusions
Rafael Brundo Uriarte
51/51
SLAC - Implementation
Compatibility only with OpenNebula I Toy Implementation I
Conclusions
Easily adapted
Rafael Brundo Uriarte
51/51
SLAC - Cloud Metrics
DTMF Cloud Computing Service Metrics Description I Recent Document (Still a Draft) I
Creation of a Model for the Definition of Metrics
I
The SLAC Metrics can be Adapted for this Model
Conclusions
Rafael Brundo Uriarte
51/51
SLAC Violation
I
Violation and Penalty are Separated Concepts
I
“Violation” Concept Flexible
I
Easy to Understand
Conclusions
Rafael Brundo Uriarte
51/51
Panoptes - Scalability
I
Designed to be scalable
I
Adapt itself
I
Experiments suggest it is scalable
I
More experiments for future works
Conclusions
Rafael Brundo Uriarte
51/51
Panoptes - Analysis of Apache Broklyn
I
Not Focused on Monitoring
I
Does Not Process the Data
Conclusions
Rafael Brundo Uriarte
51/51
Panoptes - Analysis with CSPARQL
I
Data is not Decorated (e.g. RDF)
I
Impact of Decorated Monitoring Data (Scalability)
I
Very Interesting Option
Conclusions
Rafael Brundo Uriarte
51/51