Best Dissertation Symposium: Reducing user-perceived latency in mobile applications via prefetching and caching

Slides for my PhD dissertation "Reducing user-perceived latency in mobile applications via prefetching and caching", presented at the Best Dissertation Symposium 2021 at the University of Southern California.
Presentation on YouTube: https://youtu.be/a5oTwih9rkE

Transcript

1. Reducing User-Perceived Latency
   in Mobile Applications
   via Prefetching and Caching
   Yixue Zhao
   Advisor: Nenad Medvidović
   Computing Innovation Fellow (CIFellow)
   University of Massachusetts Amherst
   

   View Slide

2. View Slide

3. View Slide

4. Latency!
   

   View Slide

5. Latency Problem
   Angry Users Sales Loss
   5
   

   View Slide

6. Main Cause: Network Transfer
   Network
   round trip
   Mobile App Remote Server
   Main cause of user-perceived latency is network transfer since
   mobile apps spend most time fetching data from remote servers.
   [Ravindranath et al. OSDI’12]
   6
   

   View Slide

7. Effective Solution!
   Problem: Annoying loading wheel L
   Solution: Prefetching and Caching
   7
   

   View Slide

8. Prefetching and Caching
   local cache
   on the phone
   immediate
   response!
   8
   

   View Slide

9. Prefetching and Caching
   9
   ?
   But how?
   

   View Slide

10. 10
    Literature
    Study
    Techniques
    Testing
    

    View Slide

11. 1. Two categories
    Literature
    Study
    Techniques
    Testing
    11
    

    View Slide

12. Two Categories
    q Content-based
    q No existing solution
    q Call for novel idea
    q History-based
    q Applicable existing solution
    q Adapt to mobile domain
    12
    

    View Slide

13. q Content-based
    q No existing solution
    q Call for novel idea
    q History-based
    q Applicable existing solution
    q Adapt to mobile domain
    13
    Techniques
    Two Categories
    

    View Slide

14. 14
    2. Is it a good idea?
    Literature
    Study
    Techniques
    Testing
    

    View Slide

15. 15
    Literature
    Study
    Techniques
    Testing
    2. Is it a good idea?
    

    View Slide

16. 16
    3a. Content-based
    (first work)
    Literature
    Study
    Techniques
    Testing
    

    View Slide

17. Content-based: PALOMA
    Key Insights
    q App’s code has all we need
    q “User think time” provides opportunities
    17
    

    View Slide

18. PALOMA at a High Level
    Code
    Analysis
    App
    Rewriting
    Optimized
    App
    18
    

    View Slide

19. PALOMA at a High Level
    19
    Automatic
    No other information
    required
    99% latency
    reduction!
    Code
    Analysis
    App
    Rewriting
    Optimized
    App
    

    View Slide

20. 20
    Literature
    Study
    Techniques
    Testing
    3b. History-based
    (practical solution)
    

    View Slide

21. History-based: HiPHarness
    21
    Existing Solutions HiPHarness
    q Large prediction models
    q Big cost
    q Privacy issues
    

    View Slide

22. History-based: HiPHarness
    22
    Existing Solutions HiPHarness
    q Large prediction models
    q Big cost
    q Privacy issues
    q Small prediction models!
    q Good training data
    q Enough training data
    

    View Slide

23. History-based: HiPHarness
    23
    Existing Solutions HiPHarness
    q Large prediction models
    q Big cost
    q Privacy issues
    q Small prediction models!
    q Good training data
    q Enough training data
    94%
    smaller!
    

    View Slide

24. HiPHarness Dataset
    q 10,000+ mobile users
    q 15+ million network requests
    q 7+ million prediction models
    24
    

    View Slide

25. q Small prediction models work great!
    q Win-win (reduce size AND improve accuracy)
    q Challenged prior conclusion
    q Re-opened this area
    HiPHarness Takeaways
    25
    Less is More!
    

    View Slide

26. 26
    Literature
    Study
    Techniques
    Testing
    4. Which is better?
    

    View Slide

27. Challenges in Testing
    q Testing prefetching and caching techniques is very
    expensive
    q Require real-user traces (i.e., “usage-based test”)
    q hire participants
    q obtain permission
    q provide guidelines
    q provide compensation
    q …
    27
    

    View Slide

28. Challenges in Testing
    q Testing prefetching and caching techniques is very
    expensive
    q Require real-user traces (i.e., “usage-based test”)
    q hire participants
    q obtain permission
    q provide guidelines
    q provide compensation
    q …
    28
    I need
    this!
    

    View Slide

29. Usage-based Test Background
    q Emerging research area in Software Testing
    (since 2018)
    q State-of-the-art research:
    test-reuse techniques
    29
    

    View Slide

30. FrUITeR at a High Level
    q Goal: find out who’s better
    q Reproduce existing techniques
    q Establish standard pipeline
    q Fair comparison in the same way
    A Framework for Evaluating UI Test Reuse
    30
    

    View Slide

31. Challenges in Practice
    Discuss with
    authors
    31
    Study
    implementation
    Modify
    implementation
    Establish
    benchmark
    Construct
    ground truths
    

    View Slide

32. Reproducibility Crisis
    32
    Common Theme
    q “It’s hard to (re)use other research tools”
    q “They don’t have documentation”
    q “It’s hard to run a research tool on my own apps”
    q …
    

    View Slide

33. 33
    FrUITeR’s Open Science Contribution
    FrUITeR’s Website:
    https://felicitia.github.io/FrUITeR/
    

    View Slide

34. v Prefetching & caching is good idea J
    encouraged more techniques
    Recap Dissertation Contributions
    34
    

    View Slide

35. v Prefetching & caching is good idea J
    encouraged more techniques
    v First content-based (PALOMA)
    inspired follow-up work
    Recap Dissertation Contributions
    35
    

    View Slide

36. v Prefetching & caching is good idea J
    encouraged more techniques
    v First content-based (PALOMA)
    inspired follow-up work
    v Make history-based practical (HiPHarness)
    challenged prior conclusion & re-opened this area
    Recap Dissertation Contributions
    36
    

    View Slide

37. v Prefetching & caching is good idea J
    encouraged more techniques
    v First content-based (PALOMA)
    inspired follow-up work
    v Make history-based practical (HiPHarness)
    challenged prior conclusion & re-opened this area
    v Test which is better (FrUITeR)
    set standards for a fair playground
    Recap Dissertation Contributions
    37
    

    View Slide

38. 4. Which is better?
    (FrUITeR)
    Dissertation & Beyond
    Mobile-App Performance
    38
    Software
    Testing
    Open
    Science
    

    View Slide

39. Dissertation Publications
    v Literature Review & Empirical Study [ASE 2018]
    v Prefetching & Caching Techniques
    [MOBILESoft 2017 SRC ICSE 2018, MOBILESoft 2021]
    v Software Testing [ESEC/FSE 2020]
    v Open Science
    [MOBILESoft 2019 Visions, ICSE 2019 Doctoral Symposium]
    39
    

    View Slide

40. Dissertation Publications
    v Literature Review & Empirical Study [ASE 2018]
    v Prefetching & Caching Techniques
    [MOBILESoft 2017 SRC ICSE 2018, MOBILESoft 2021]
    v Software Testing [ESEC/FSE 2020]
    v Open Science
    [MOBILESoft 2019 Visions, ICSE 2019 Doctoral Symposium]
    40
    https://tinyurl.com/yixuedissertation https://tinyurl.com/yixuetalks
    

    View Slide

41. Thank you!
    41
    

    View Slide