Tool-based Interactive Parallelization: A Case Study

Takeaway 1: Leveraging future hardware requires software comprehension
Takeaway 2: Parallelization requires comprehension of behavior and structure
Takeaway 3: There is an abstraction gap between software systems and analysis tools

Transcript

1. TOOL-BASED INTERACTIVE SOFTWARE TOOL-BASED INTERACTIVE SOFTWARE PARALLELIZATION: A CASE STUDY PARALLELIZATION: A CASE STUDY A. WILHELM A. WILHELM, F. CAKARIC, T. SCHÜLE, M. GERNDT, ICSE-SEIP 2018 , F. CAKARIC, T. SCHÜLE, M. GERNDT, ICSE-SEIP 2018 1
2. 50 YEARS OF SOFTWARE ENGINEERING 50 YEARS OF SOFTWARE ENGINEERING A real success story 2 . 1
3. RIDING ON MOORE'S LAW RIDING ON MOORE'S LAW 2 . 2
4. WE WILL HAVE A WILD FUTURE WE WILL HAVE A WILD FUTURE https://www.chiprebel.com/apple-a11-bionic 2 . 3
5. KNOW YOUR DEPENDENCIES KNOW YOUR DEPENDENCIES 2 . 4
6. LISTEN TO JOHN L. HENNESEY, ET.AL, NOT ME LISTEN TO JOHN L. HENNESEY, ET.AL, NOT ME https://www.acm.org/turing-award-50/video/moores-law-is-really-dead 2 . 5
7. LEGACY CODE MEETS MULTI-CORE LEGACY CODE MEETS MULTI-CORE https://www.siemens.com 3 . 1
8. MARGARET HAMILTON, ICSE 2018 MARGARET HAMILTON, ICSE 2018 Systems are asynchronous, distributed and event driven in nature. This should be re ected in the language to de ne them and the tools to build them [...] 3 . 2
9. OUR MODUS OPERANDI OUR MODUS OPERANDI 1. Identify concurrency 2. Redesign the software architecture 3. Parallelize the code 4. Validate and verify 5. Analyze the runtime 3 . 3
10. HIGH-LEVEL PROBLEM HIGH-LEVEL PROBLEM 3 . 4
11. LOW-LEVEL PROBLEM LOW-LEVEL PROBLEM 3 . 5
12. SOLUTION A: HIRE NINJAS SOLUTION A: HIRE NINJAS https://github.com/codewarz-ninja 3 . 6
13. SOLUTION B: USE TOOLS SOLUTION B: USE TOOLS 3 . 7
14. PARALLELIZATION TOOLS PARALLELIZATION TOOLS Very useful for regular code Operate on low abstraction levels 3 . 8
15. ARCHITECTURE ANALYSIS TOOLS ARCHITECTURE ANALYSIS TOOLS Help to understand and evaluate Neglect dynamic aspects 3 . 9
16. COMPREHENSION IS ESSENTIAL COMPREHENSION IS ESSENTIAL (of the structure and the behavior) 3 . 10
17. 4 . 1
18. WORKFLOW WORKFLOW 4 . 2
19. I WILL NOT TALK ABOUT I WILL NOT TALK ABOUT 4 . 3
20. CPPCHECK CODE ANALYSIS CPPCHECK CODE ANALYSIS 4 . 4
21. DEPENDENCY ANALYSIS DEPENDENCY ANALYSIS https://youtu.be/dbBjWdcw--k 4 . 5
22. ARCHITECTURE ANALYSIS ARCHITECTURE ANALYSIS https://youtu.be/4yCJ7FIKsCg 4 . 6
23. SOFTWARE ARCHITECTURE RECONSTRUCTION, SOFTWARE ARCHITECTURE RECONSTRUCTION, ANOTHER TRY? ANOTHER TRY? 4 . 7
24. SO, YOU USE DIRECTORIES? SO, YOU USE DIRECTORIES? controller := inFile("/src/controller/.**") 4 . 8
25. OR NAMESPACES? OR NAMESPACES? logic := namespace("log") 4 . 9
26. SHARED LIBRARIES ANYBODY? SHARED LIBRARIES ANYBODY? libs := image("/3dparty/.*.so") 4 . 10
27. ALL TOGETHER? ALL TOGETHER? operations := libs || function("operate.*") && !logic 4 . 11
28. PARCEIVE HELPS TO... PARCEIVE HELPS TO... 1. Identify concurrency +++ 2. Redesign the software architecture ++ 3. Parallelize the code - 4. Validate and verify + 5. Analyze the runtime -+ 4 . 12
29. SOME FINAL THOUGHTS SOME FINAL THOUGHTS 5 . 1
30. WHY BINARY ANALYSIS? WHY BINARY ANALYSIS? No source code needed Optimizations and generics Language independent Slowdown Complex deployments 5 . 2
31. THE FACTS THE FACTS Biggest software applied: >10M loc Supported plattforms: Linux, Windows Supported languages: C/C++ Current state: Working prototype 5 . 3
32. WE STRIVE FOR COLLABORATIONS WE STRIVE FOR COLLABORATIONS I brought some business cards 5 . 4
33. QEUSTIOSN? QEUSTIOSN? 5 . 5