Software Architecture Metrics Hany Ammar, Mark Shereshevsky, Ali Mili, Walid Rabie and Nicholay Gradetsky Lane Department of Computer Science & Electrical Engineering (smark, amili, rabie,

Initiative Title: Architectural Level Software Metrics '01 Architectural Level Software Metrics '01 FY2001 University SOFTWARE INITIATIVE PROPOSAL for the NASA SOFTWARE IV&V FACILITY

Outline Project Overview Hub Control Software (HCS) case study Definition of the entropy based metrics Definition of the Quantitative Factors Informational Integrity Coefficient (IIC) Conclusions (achievements) Papers submitted Work In Progress Future work (2 nd year plans)

Project Overview Developing a class of architectural metrics that quantify information flow in the architecture. Different metrics reflect: Coupling vs. cohesion; Data flow vs. control flow; Static measurement vs. dynamic measurement; Introducing a group of quantitative factors: mathematically defined functions, that are relevant with respect to qualitative attributes of the architecture Establishing analytical relationship between the metrics and the quantitative factors Automating computation of the metrics

Product Artifacts Architecture, Design, Code, External Attributes: Maintainability,Testability, Reliability, Computable Metrics: Coupling, cohesion, complexity, etc. Error Propagation, ChangePropagation, Requirements Propagation impacts computed for Analytical Definitions Mathematical Formula have Historical Data Measurement Values produce Validation Analytical Derivations Quantified McCall's model of software quality [McCall 1977], Three-level hierarchy of software attributes Project Overview

Project Overview McCall's model of software quality in UML

Hub Control Software (HCS) Case Study International Space Station

General HCS-ITCS scheme HCS ITCS SCITCS FRITCS LRITCS PPA mon PFMC LT PFMC MT Scheduler State manger CMD Queue O/P CMD Queue N3-1 Data Access RPCM N3-2 Data Access Other HCS sub-systems

HCS case study: Hierarchical Architecture

Definition of entropy based metrics n Definition 1: Coupling: The coupling from A to B, denoted Coupling(A,B), is the entropy of random variable V A  B. Cohesion: The cohesion of A is defined as coupling from A to A. Ensemble CNTL, DATA  E p is the probability distribution of E. V A  B is a random variable defined by E and p S CNTL AB DATA

HCS case study : ITCS Static Control Matrix ScitcsFritcsPfmc mtpfmc ltlritcsppa mon Scitcs Fritcs Pfmc mt1220 pfmc lt1202 lritcs ppa mon

HCS case study : HCS Static Control Matrix (LEVEL 1) itcsN3queuesrpcmschstate man itcs N Queues rpcm sch state man

Information Theoretic metrics Method of data collection Aspect of inform. exchange Static (syntactic analysis, Comprehensive statistics) Dynamic (running actual instances of the architecture) Control Entropy of set of methods, with statically computed probabilities Entropy of set of methods, with dynamically computed probabilities Data Entropy of set of parameter values, with statically computed probabilities Entropy of set of parameter values, with dynamically computed probabilities

HCS case study : ITCS Dynamic Control Matrix ScitcsFritcsPfmc mtpfmc ltlritcsppa mon Scitcs Fritcs Pfmc mt pfmc lt lritcs ppa mon

HCS case study : HCS Dynamic Control Matrix (LEVEL 1) itcsN3queuesrpcmschstate man itcs N Queues rpcm sch state man

Definition of the Quantitative Factors n Quantitative Factors bridge the gap between qualitative attributes (which we want to assess) and computable metrics (which we can measure). n Some Quantitative Factors: Error Propagation Change Propagation Requirements propagation

Definition of the Quantitative Factors: error propagation  B (x) is the state of B resulting from invoking B on input x. Suppose, due to a fault in A, the information transmitted from A to B is corrupted from the expected x to some erroneous x`.  The error propagation (EP) is a measure of the likelihood that an error in the message sent by A will propagate into B. S X AB x x` x OR Error

Definition of the Quantitative Factors: error propagation Continued n Definition 2: Error Propagation: The error propagation from A to B is defined as the following conditional probability: EP(A,B) = Prob( B (x)  B (x’) | x  x’). This equation can be reduced to: EP(A,B) = 1 – exp( – Coupling(A, B)) + exp( – Cohesion(B)).

Definition of the Quantitative Factors: error propagation Continued From the analysis of the above equation:  The higher the coupling from A to B – the higher is the error propagation from A to B.  The lower the cohesion of component B – the higher is the error propagation from A to B. These observations, consistent with intuition, provide an analytical validation of our metrics. Our equation gives these relationships (which are not surprising by themselves) precise analytical form.

Informational Integrity Coefficient (IIC) n One of architecture quality guidelines is to maximize cohesion and minimize coupling of components. n We introduce a quantitative characteristic, which we call the informational integrity coefficient of an architecture which assesses its conformance to this guideline. n The coefficient is computed based on the information flow matrix. IIC =

HCS case study : HCS-ITCS (flattened) Static Control Matrix IIC = 0.749

HCS case study : ITCS Static Control Matrix ScitcsFritcsPfmc mtpfmc ltlritcsppa mon Scitcs Fritcs Pfmc mt1220 pfmc lt1202 lritcs ppa mon IIC = 0.795

HCS case study : HCS Static Control Matrix (LEVEL 1) itcsN3queuesrpcmschstate man itcs N Queues rpcm sch state man IIC = 0.861

IIC = HCS case study : HCS-ITCS (flattened) Dynamic Control Matrix

HCS case study : ITCS Dynamic Control Matrix ScitcsFritcsPfmc mtpfmc ltlritcsppa mon Scitcs Fritcs Pfmc mt pfmc lt lritcs ppa mon IIC = 0.884

HCS case study : HCS Dynamic Control Matrix (LEVEL 1) itcsN3queuesrpcmschstate man itcs N Queues rpcm sch state man IIC = 0.887

HCS case study : IIC Comparison Flattened Architecture Vs Hierarchical Architecture ArchitectureStaticDynamic HCS-ITCS (flattened) HCS_ITCS (Level 1) ITCS (Level 0) N3 (Level 0)11 Queues (Level 0)11 n The flattened architecture shows a lower IIC than the hierarchical architecture, this proves that the hierarchical architecture is more cohesive and less coupled

Conclusions (achievements) We defined coupling and cohesion metrics for hierarchical architectures based on the information theory. We introduced a normalized measure, the informational integrity coefficient (IIC) of an architecture based on the computed metrics We derived the relationship between the coupling/cohesion metrics and the error propagation factor between components.

Papers submitted Conferences Annual International Computer Software and Applications Conference (COMPSAC 2001) Title: "Information-Theoretic Metrics for Software Architectures“ – Status: Accepted for publication Automated Software Engineering Conference (ASE 2001) Title: "Quantifying Architectural Attributes of UML Specifications: A Framework and Its Automation" – Status: Submitted June 29, 2001 Journals Automated Software Engineering Journal Title: "Quantifying Software Architectures: Definitions and Automation Plans“ – Status: submitted: May 11, 2001 Transactions on Software Engineering Journal Title: "Information Theoretic Metrics for Software Architectures" – Status: submitted: April, 2001

Future Work Deriving a relationship between our coupling/cohesion metrics and the change propagation, Developing an automation methodology for computing the metrics given a UML description of the architecture, Validating the metrics using a NASA CASE Study

Canonical Architectural Style Automation support for metrics computation Software System Conceptual view (RoseRT-UML) Static Metrics Quantitative factors (EP, CP) In progress Planned Has a view Other views (UML, Rapide) Dynamic Metrics Transformation Information Integrity Coefficient