Rob Baarda - Are Real Test Metrics Predictive for the Future?
Download as PPT, PDF•0 likes•1,091 views
The document discusses the use of real metrics in software testing and their potential predictive capabilities for project management. It emphasizes the importance of accurate definitions and the collection of relevant data for effective benchmarking and assessing testing processes, focusing on aspects like defect detection, removal efficiency, and productivity. Various models and statistical analyses are suggested for improving the estimation of testing efforts and outcomes in future projects.
Rob Baarda - Are Real Test Metrics Predictive for the Future?
- 1. Are real metrics predictive for the future? [email protected]
- 2. Agenda • Introduction • Objectives? GQM! • Real metrics • Considerations • Use real metrics in your future
- 3. Which test metrics? Test basis Test object Test Execution Defects Repair Production Specifying test cases / scripts Test cases/ scripts Test Process Size test basis Size test object # defects in test object # defects in production For each process: # hours effort lead time # test cases # = number of # defects in test basis # repair rounds
- 4. Deductible metrics • Effort estimation = # hours /size (FP, KLOC) • Productivity = # test cases / # hours • Efficiency = # defects / (# hours or # test cases) > Specification > Test execution > Retest of repaired defects • DDP Defect Detection Percentage (Europe) DRE Defect Removal Efficiency (USA) • Defect injection rate for rework • Damage prevented in €? HOW to get data? HOW to organize?
- 5. Dutch test metrics experiences • Dutch initiative to gather test metrics • Parties involved >NESMA Netherlands Software Metrics Association >Testnet Dutch Testing community >LaQuSO Laboratory for Software Quality Universities Eindhoven & Nijmegen Approach Goal Question Metrics (GQM)
- 6. 6 Goals 1. Test manager Support in planning and controlling the testing project 2. Organization Benchmark around 1. Test process 2. Test products 3. IT-products To improve test process, IT process
- 7. 7 Some Questions • Test manager > Number of test cases needed for my project > What percentage of the project team should be allocated to testing > How many retests are executed • Organization Benchmark > What is the defect detection & removal efficiency (at what phase) > What test coverage do I need to ensure adequate testing > How many defects does development insert when repair others
- 8. What we have - Structure Project Test project Test activity Incidents/ Defects Project activity Be careful using the data Lack of statistical evidence
- 9. Feedback example test effort % test effort / project effort % test effort / project effort Project effort (man-day) 0 10 20 30 40 50 60 0 500 1000 1500 2000 2500 3000 Be careful using the data Lack of statistical evidence
- 10. Feedback example test productivity Test productivity Size in Function Points # Test hours / FP 0 2 4 6 8 10 12 14 16 18 0 100 200 300 400 500 600 Be careful using the data Lack of statistical evidence
- 11. Feedback example defects per fp 0 0,2 0,4 0,6 0,8 1 1,2 0 100 200 300 400 500 600 700 800 900 Number of defects / function point Size in Function Points # defects / fp Be careful using the data Lack of statistical evidence
- 12. Defect Detection Percentage Defect Removal Efficiency DDP (%) 84 85 86 87 88 89 90 91 92 0 100 200 300 400 500 600 700 800 900 1000 Size in SKLOC Defect Detection Percentage Defect Removal Efficiency Be careful using the data Lack of statistical evidence
- 13. Processes around metrics • Collection in a project > Embedded in daily work > Weekly summarisation > Sanity checks > Cost: about 2% project budget • Distribution • For a benchmark on the level of: > Project releases > Organisation > Country > International: www.ISBSG.org International Software Benchmarking Standards Group
- 14. Some Considerations for future use 1. Accuracy of definitions 2. Number of types of defects 3. Is a batch test case the same as an online test case? 4. Only testing of functionality or also security, performance, usability 5. How to include regression testing? 6. Measure personal productivity? 7. Predictive value average (mean), median, standard deviation, correlations with? Prediction model needed?
- 15. 10 similar projects Project Func Design Construct System test Function Points FD-hrs per fp Constr-hrs per fp Systemtest- hrs per fp 1 285 465 183 95 3,00 4,89 1,93 2 631 1847 694 305 2,07 6,06 2,28 3 599 845 540 197 3,04 4,29 2,74 4 159 496 185 57 2,79 8,70 3,25 5 81,5 1057 306,5 93 0,88 11,37 3,30 6 416 1017 281,5 80 5,20 12,71 3,52 7 528 1069 605 137 3,85 7,80 4,42 8 566 3118 756 176 3,22 17,72 4,30 9 848 5834 1776 265 3,20 22,02 6,70 10 508 4666 2204 285 1,78 16,37 7,73 Average 4,0
- 16. Standard deviation From wikipedia Standard deviation = the mean root square (RMS) deviation of the values from their mean(=average): σ
- 17. Some statistics for ST hours/fp • Average = 4.0 • Standard deviation = 1.8 As predictor: 68% will be between 2.2 and 5.8 • Does not really help as prediction
- 18. Additional info Project Func Design Construct System test Function Points FD-hrs per fp Constr-hrs per fp Systemtest- hrs per fp Grafical system 1 285 465 183 95 3,00 4,89 1,93 0 2 631 1847 694 305 2,07 6,06 2,28 0 3 599 845 540 197 3,04 4,29 2,74 0 4 159 496 185 57 2,79 8,70 3,25 0 5 81,5 1057 306,5 93 0,88 11,37 3,30 0 6 416 1017 281,5 80 5,20 12,71 3,52 0 7 528 1069 605 137 3,85 7,80 4,42 0 8 566 3118 756 176 3,22 17,72 4,30 1 9 848 5834 1776 265 3,20 22,02 6,70 1 10 508 4666 2204 285 1,78 16,37 7,73 1
- 19. And now • Non-GIS > Average = 3.1 > Standard deviation = 0.8 > As predictor: 68% within 2.3 and 3.9 • GIS > Average = 6.2 > Standard deviation = 1.4 > As predictor: 68% within 4.8 and 7.6 • Overall > Average = 4.0 > Standard deviation = 1.8 > As predictor: 68% within 2.2 and 5.8
- 20. Use metrics in your future 1. Starting point for project > Use estimation model, mostly linear, use categories Small, Middle, Large > Use “common” metrics Possible source: Chapter 11 of TMap® Next book 1. Look at your real project data, consistent with prediction? > Yes: GO TO End 1. Find the major factor influencing 2. Adapt your 1. Estimation model 2. Metrics 3. GO TO 2
- 21. Wrap up • Metrics are possible • Useful to predict • Linear model needs localized fine tuning Test Metrics can predict your future!
Editor's Notes
- #21: Extra Voorbeeld: schatting bij testline Guido N. : 3 categorieën, ronde 1: extra categorie (ES), ronde2: staffelregels voor grotere aantallen (= meenemen inleereffect)