Universität Karlsruhe (TH) Research University • founded 1825
Reengineering of Software-Components for Prediction of Extra-Functional Properties Reengineering Component Behaviour Focus: Performance Prediction
Klaus Krogmann (
[email protected]) Chair Software Design and Quality Institute for Program Structures and Data Organization (IPD) Faculty of Informatics, Universität Karlsruhe (TH)
Performance Prediction
Performance meta-model of a component based software architecture
Krogmann: Reengineering Software-Component Behaviour
Performance data ▪ Execution time ▪ Response time
07-05-13
2
Component Performance
[Becker2006a] Krogmann: Reengineering Software-Component Behaviour
07-05-13
3
Black-Box vs. Grey-Box Execution Time of a()? b() 2ms c() 3ms ComponentB
a() ?ms ComponentA
d() 5ms ComponentC
Krogmann: Reengineering Software-Component Behaviour
07-05-13
4
Overview Motivation Example: Desired outcome Assumptions Idea: Process / How Conclusion
Photo: pixelquelle.de
▪ ▪ ▪ ▪ ▪
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
5
Performance Prediction ▪ A pure black-box view of components is not sufficient to predict component performance ▪ How to gain a grey-box view of existing components? – Control flow – Data flow – Abstraction level: information influencing component performance
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
6
Service Effect Specification getListWithLittleEntropy
public List getListWithLittleEntropy (List listToSort, int count) { while(mode) {
// some simple internal action for(int x = 0; x < count; x++) { listToSort.add(new Integer(x)); }
if(count > 100) { //external call: collectionComponent.sort(listToSort); }
CollectionComponent.sort()
//external call: mode = collectionComponent. isEntropyLessThan(listToSort, count);
CollectionComponent.isEntropyLessThan()
} return listToSort; }
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
7
Expected Result of Analysis (1) Resource Demanding SEFF ▪ Parametric dependencies / variable usage ▪ Resource usages ▪ Stochastic expressions ▪ Allows reasoning on – Functional dependencies – “QoS-Attributes” of component / SEFF – Execution probabilities Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
8
Expected Result of Analysis (2) ▪ Parametric dependencies / variable usage – In: Input parameter – Used for • • • • • •
Internal actions Branching (branch selection) External calls (variable usage) Loops (# iterations) Time between resource acquisition and release Return values
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
9
Expected Result of Analysis (3) ▪ Parametric resource usages – CPU (cycles) – Used memory – Network
▪ Abstractions / Approximations – not the “real dependency”
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
10
Idea ▪ Combination of – Static source code analysis for control flow structure – Analysis of components at runtime
▪ Use machine learning (i.e. genetic algorithms) to get the required the abstraction level
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
11
Assumptions (1) ▪
Given – Source code – Component boundaries and component interfaces (including roles) (Æ predefined what internal and external calls of a SEFF are)
▪ ▪
Assumption: Source code can be compiled All external services can either be simulated or mocked (Æ related work exists)
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
12
Assumptions (2) ▪
Testing approach: – Components are monitored at runtime – Not the whole input space can be covered within finite time – Need to find adequate test data Æ Assumed that such test data is given • Test data is independent from a specific usage profile
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
13
Intended Process
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
14
Genetic Algorithm ▪ In: – Measuring data from tracing – Optimization function • Coverage with measuring data of tracing • Complexity of expression evaluation
– Initial Building Blocks
▪ Out – Equation (parameterized)
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
15
Initial Building Blocks
void process(int a){ doIt1(); do { doIt2(); a++; } while (a < 10) } [Overall_runtime] = [method_init] + [const1] + ( [while(a)] * [const2] );
sequence
nesting
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
16
Example: Internal Action int f(int p) { for(int i = 0; i < 1000; i++) { p++; p += 1; //do some useless calculations p -= 1; } return p; } ▪ Functional: f(p): p + 1000 ▪ Execution time: f(p): O(1) J constant ▪ Both values are relevant for SEFF annotations Krogmann: Reengineering Software-Component Behaviour
07-05-13
17
Limitations ▪ Reconstruction of concurrency is not planned ▪ The monitored behaviour usually depends on the chosen test cases to monitor the source code at run-time. ▪ Instrumentation of source code for monitoring purposes can create very big amounts of measuring data.
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
18
Conclusion ▪ Approach to reengineer component internals to enable performance prediction ▪ Combination of static and dynamic reengineering ▪ Use of genetic algorithms for finding abstractions of parametric dependencies
Motivation ` Example: Outcome ` Idea: Process ` Conclusion Conclusion Krogmann: Reengineering Software-Component Behaviour
07-05-13
19
