Architecting an Evolvable System by Iterative Object Process Modeling

1
Architecting an Evolvable System by Iterative
Object Process Modeling

• Presented by Ashirul Mubin
• The University of Alabama Graduate School
• Contributors Daniel Ray, Math Computer
  Science, University of Virginia
• Rezwanur Rahman, Aerospace Engineering, The
  University of Alabama

2
Problem Scope

• Difficult to foresee future emerging requirements
  at the time of large system development
• Unable to properly meet the newly emerged
  requirements
• makes the system to gradually lose its value
• The users become less satisfied customers do
  not get proper services
• Eventually, the system becomes outdated
• Very costly to replace the entire system

3
Objectives

• Enable newly developed large system to sense
  current state of evolution
• Identify and track the emerging new requirement
  changes in specific areas of the system workflow
• Deduce the inception time of the next
  evolutionary phase from system feedback
• Provide as much information/data as necessary to
  apply the changes to the system

4
Redesign of Systems Comparison
Evolving smaller system by replacements
Evolution of Camry Over last 20 years
Evolution of Sony TV Over last 20 years
Evolution of large system by iterative changes in
system requirements design
Evolve by iteration
Evolve by replacement
Evolve by replacement
Evolve by replacement
5
User Satisfaction Index (USI)

• To quantify the system value with level of
  satisfaction of the users (and affiliates).
• It may also quantify the service-quality to its
  customers
• Psychological factors are also involved
• A system architect has the freedom to define a
  system specific USI
• that will most closely quantify its system value
• that will also give correct index of users
  overall experiences
• USI indicates the value for an instant of time
  (time stamped)

6
Examples of Quantifying Factors

• Development activity log
• Tracking of update/change history
• Request Inter-arrival time
• Service (wait) time
• System usage activity log
• Data flow control flow patterns
• Component or Feature ranking of
• The existing features
• The newly requested features
• Collecting user experiences suggestions
• On-page feedback
• Periodic surveys
• Changes in system state
• meta-data meta-model

7
Behavior of a Traditional System
8
Behavior of an Evolvable System
9
USI Metrics for Analyzing System Behavior
the general purpose is to guide the Analyzer
throughout an evolvable cycle.
10
Evolutionary Factors
the purpose is to provide Decision Support Data
(DSD) to the analyzer but it will vary widely
from system-to-system.
11
Iterative Architecture
The wrapper system coordinates three basic stages
that iterate through an evolutionary cycle at the
inception of evolving factors.
12
Stage 1 Capturing System Behavior

• Identify system/environmental parameters
• place probing points in system workflow to
  collect behavioral data into probing station
• Workflow activity pattern
• Change request history
• Request inter-arrival time
• Feature ranking
• Manual feedback/survey
• Establish a system measurement profile
  (behavioral data) from the deduced system metrics
  to aid computing inception of evolutionary phase,
  as well as any needed changes in the system

13
Stage 2 Updating System State

• Collected behavioral data is fed into Evolution
  Policy Analyzer that updates system states
  (system meta-data and system meta-model)
• The purpose of the analyzer is to accurately
  predict the inception of evolutionary phases by
  applying appropriate statistical models
• Based on non-parametric Kernel Density Estimation
  to predict future probable inter-arrival times
• Hypothesis testing for more detail rigorous
  testing
• Comparative Histograms to show close match of the
  prediction with that of the actual data
• Running Models from continuously changing data at
  each iteration
• At each iteration, look for any Transient
  Indicators
• ? whether steady state is reached after sometime

14
Stage 2 (cont.) Statistical Models for the
Analyzer

• The histograms from given data set and estimated
  from Kernel Density Estimation technique are
  highly correlated. This may help to predict the
  efficiency of the system based on Monte Carlo
  Simulation.

November 09
THE UNIVERSITY OF ALABAMA, TUSCALOOSA, AL 35487
14
15
Stage 3 Applying New Changes

• When the Evolution Policy Analyzer signals the
  inception of a new evolutionary phase, the
  wrapper system (probing station, meta-data
  meta-model) provides necessary specifications for
  updating the target system (by service
  orientation)
• By selecting appropriate tools methodologies
  (such as SaaS , integrated architectural approach
  4 or reflective arch 5.
• Short, gradual incremental updates
• Adjust probing points in the revised workflow
• Keep probes live, attached to appropriate places
  in the workflow
• Continue collecting data for the upcoming cycles
• Progressively maintain an evolvable system for
  its extended life cycle

16
Modeling the system using OPCAT
Developed by MIT research affiliates, OPCAT is a
system modeling tool that uses Object Process
Methodology
17
General Observation from Empirical Data

• Calculation of USI by quantifying factors
• System development activity log
• Update/change tracking history (inter-arrival
  times)
• System state at inception time (previous USI)
• Promptness of addressing the requested changes
• Service time (wait time)
• User experiences
• by Surveys On-page feedbacks
• Key indicators
• Request inter-arrival times
• Frequency of update requests
• Collective USI

18
Applying Evolutionary Factors
19
Conclusion

• Significant benefits can be obtained from a large
  evolvable system
• Clear insight of the system
• Increased control over maintaining upgrading
  the system
• Much lower maintenance cost
• Progressively maintain satisfactory system
  outcomes
• In these days of technological advancements, the
  system architects and developers have much wider
  freedom for developing evolvable systems
• Future Considerations
• Automated ways to update system objects,
  processes and codes needs to be investigated
  further (tools, like OPCAT)
• An in-depth study of environmental and
  psychological impact on USI changes for each
  iteration will help refine related metrics
  identify effective evolutionary factors
• Applying further statistical analyses on the
  accumulated historical data will direct more
  accurate prediction of evolutionary cycles

20
Questions?