Definition%20of%20Metric%20Dependencies%20for%20Monitoring%20the%20Impact%20of%20Quality%20of%20Services%20on%20Quality%20of%20Processes

1
Definition of Metric Dependenciesfor Monitoring
the Impact of Quality of Serviceson Quality of
Processes

• Christian Mayerl, Kai Moritz Hüner, Jens-Uwe
  Gaspar, Christof Momm, Sebastian Abeck
• Cooperation Management, Institute of
  Telematics,Faculty of Computer Science,
  Universität Karlsruhe (TH)76128 Karlsruhe,
  Germany
• mayerl huener gaspar momm
  abeck_at_cm-tm.uka.de

2
INTRODUCTION Overview

• Introduction
• Related Work
• Metric Dependency Model
• Elements for modeling metric dependencies
• Formalism for calculating metric dependencies
• Example Performance Metric Dependencies
• Dependency model of response time
• Application of response time metrics
• Monitoring Architecture
• Components processing metric definitions
• Implementation experiences and evaluation
• Conclusion and Outlook

3
INTRODUCTION Scenario and Questions

• Which elements needed to define dependencies
  between metrics
• related to processes, invoked services and
  components implementing services?
• How to define metric dependencies to allow
• mapping of metric dependency definitions to an
  monitoring architecture?

4
RELATED WORK Overview

• Basics of quality definitions
• UML Profile for Modeling Quality of Service
• basic terminology and structure, no calculable
  quality dependencies
• Quality of Web Services
• raw and aggregated metrics, no dependencies
  between services and processes
• Metric dependencies
• Business-driven IT Management (BDIM)
• optimize SLOs from business perspective, not
  monitoring dependencies at runtime
• Dependencies between service and process metrics
• e-business processes based on Web services, no
  management function to monitor Quality of Service
  (QoS) and to calculate the impact on Quality of
  Process (QoP)
• Management architectures and applications
• Common Information Model (CIM) Metrics Model
• metrics related to functional units, no
  definition of calculable metric dependencies

5
RELATED WORK Common Information Model (CIM)
Metrics Model

• Modeling of runtime management information
• Metric dependencies associated with relation
  CIM_SubUoW between units of work

6
METRIC DEPENDENCY MODEL Assumptions and
Requirements

• Assumptions
• focus on one process activity and one invoked
  service operation
• process implemented with Business Process
  Execution Language (BPEL)
• services described with Web Service Description
  Language (WSDL)
• services implemented by component-based
  applications
• functional dependencies between services and
  processes are well-known
• Requirements
• provider monitoring Quality of Services (QoS)
  should be able to estimate impacts on Quality of
  Processes (QoP)
• formal model of metric dependencies needed to be
  calculated
• mapping on existing standard management
  information models (e. g. Common Information
  Model, CIM)

7
METRIC DEPENDENCY MODEL Metrics and Metric
Dependencies
Process Layer
Metric Dependencies
Service Layer
Application Layer
8
METRIC DEPENDENCY MODEL Formalizing Metric
Dependencies
9
EXAMPLE PERFORMANCE METRIC DEPENDENCIESIntroduc
ed Example Scenario
ProcessLogic
Business Process Activity ToRProcess.CollectStuden
tInformation
functionaldependency
Service
Service Operation StudentDBService.getCompleteSet
functionaldependency
ApplicationLogic
Component FunctionStudentDBAccess.getCompleteSet

• Process activity invokes a service operation
• Service implemented by application component
  functions

10
EXAMPLE PERFORMANCE METRIC DEPENDENCIESMetric
Dependency Pattern for Performance
requestsent
processingstarted
responsesent
responsereceived
requestreceived
ProcessLayer
ServiceLayer
ApplicationLayer

• Metric dependency patterns based on existing
  approaches
• Pattern applied to specific SOA by assigning
  concrete architectural elements to each metric

11
EXAMPLE PERFORMANCE METRIC DEPENDENCIESModel
for StudentDBService and ToRProcess
12
EXAMPLE PERFORMANCE METRIC DEPENDENCIESSimple
Metric Dependency Function

• Assumptions
• Dependency functions

13
MONITORING ARCHITECTURE Component Overview
14
MONITORING ARCHITECTURE Component Details

• AgreementMonitor
• MetricAggregator
• calculating metric values based on functional and
  metric dependencies
• MetricComparator
• comparing SLOs with calculated metric values
• MetricCollector
• collection of metric values and transformation to
  management information standards (e. g. CIM)
• Others
• Storage
• Archiving of measured values
• Associating values with functional and metric
  entities
• Instrumentation
• manageability capabilities of functional elements
• provision of sufficient management information

15
IMPLEMENTATION EXPERIENCES Demonstrator

• 10.000 requests for collecting and aggregating
  metric values

16
IMPLEMENTATION EXPERIENCES Evaluation of
MetricCollector and MetricAggregator

• Useful medians
• 141 milliseconds for MetricCollector
• 437 milliseconds for MetricAggregator

17
IMPLEMENTATION EXPERIENCES Evaluation of Metric
Dependencies (Component, Service, Process)
milliseconds
18
IMPLEMENTATION EXPERIENCES EVALUATIONSorted
Measurement Data
milliseconds
19
IMPLEMENTATION EXPERIENCES EVALUATIONCorrelatio
n of Measurement Data
Service
Component
Process
Process
Component
Service
20
CONCLUSION AND OUTLOOK

• Conclusion
• Modeling metric dependencies between elements of
  a SOA
• Formalism to calculate metric dependencies
• Example response time dependencies
• Monitoring quality impacts at run time
• Development of a monitoring architecture
• Evaluation of monitoring metric dependencies
• Outlook
• Definition of a UML profile to model metric
  dependencies
• Support for development of manageable SOA
  elements
• Integrated development of SOA and manageability
  capabilities
• Extensions of the management instrumentations
• Development of adequate management interfaces
• Provision of management information for
  calculating metric dependencies