SATURN 2015 - Never again offline

Transcript

1. © Fraunhofer IESE 1 Never again offline?!? Experiences on the outstanding role of data in a large-scale mobile app ecosystem Dr. Matthias Naab, Fraunhofer IESE Dr. Ralf Carbon, John Deere ETIC Susanne Braun, Fraunhofer IESE April 29, 2015 SATURN 2015, Baltimore, USA
2. © Fraunhofer IESE 2
3. © Fraunhofer IESE 3
4. © Fraunhofer IESE 4 John Deere Ecosystem My John Deere
5. © Fraunhofer IESE 5 Introducing Mobile to the John Deere Ecosystem ◼ One app vs. many apps ◼ Native vs. hybrid or web ◼ Online vs. Offline ◼ Proprietary vs. standard data ◼ … My John Deere
6. © Fraunhofer IESE 6 Collaboration Project ◼ Duration: since Q4 2013 ◼ Several phases and scope extensions ◼ Team ◼ ~ 6 full time persons ◼ Approach ◼ UX / Arch / Dev / QA ◼ 2 week iterations ◼ Testing with real end customers since summer 2014 ◼ Continuously feeding concepts into production version ◼ Re-newing the code base after ~6 months
7. © Fraunhofer IESE 7 Where to find Data in a Typical Architecture Document?
8. © Fraunhofer IESE 8 Where is Data Really???
9. © Fraunhofer IESE 9 What does this mean for architecting?
10. © Fraunhofer IESE 10 We need an App … Architecting it should be Simple! 1
11. © Fraunhofer IESE 11 We need an App … Architecting it should be Simple! 1 JobManager iOS App DB: SQLite CoreData UI MVC iOS Data Modeling
12. © Fraunhofer IESE 12 Multiple Users Work on Shared Enterprise Data 2
13. © Fraunhofer IESE 13 Multiple Users Work on Shared Enterprise Data 2 Device 2: iPhone Device 1: iPad JobManager iOS App MyJobs iOS App • Data exchange via backend Cloud Backend DB: MySQL Hibernate Logic Mobile Services HTTP REST JSON
14. © Fraunhofer IESE 14 Apps Work on Shared Data 3
15. © Fraunhofer IESE 15 Apps Work on Shared Data 3 • iOS <=7 : No shared data JobManager iOS App DB: SQLite CoreData UI MyResources iOS App DB: SQLite CoreData UI ?
16. © Fraunhofer IESE 16 Apps Work on Shared Data 3 • Data exchange via backend • On app switch, update data JobManager iOS App MyResources iOS App Cloud Backend DB: MySQL Hibernate Logic Mobile Services HTTP REST JSON
17. © Fraunhofer IESE 17 Users Have to Work Offline 4
18. © Fraunhofer IESE 18 Users Have to Work Offline 4 • Custom sync mechanism • Sync service and engine • UUID + RevisionID + SyncState • Conflict detection + resolution • Treatment of deletion • → High effort! Cloud Backend DB: MySQL Hibernate Logic Mobile Services Device 1: iPad JobManager iOS App DB: SQLite CoreData Sync Engine UI Sync Device 2: iPhone MyJobs iOS App DB: SQLite CoreData Sync Engine UI job { "id": “UUID", "revNr": "42“, "syncState": “UPDATED" }
19. © Fraunhofer IESE 19 Data has Different Characteristics 5 Field Machine Person Job Assignment GPS Coordinate GPS Coordinate • Data classification • Data modeling rules (e.g. transactional accesses master, usage of aggregates) • Life-cycle of objects (deletion) Master Data Transactional Data Entities Value Objects
20. © Fraunhofer IESE 20 JD Operates Multiple Tenants on the Backend 6 job { "id": “UUID", “tenantId": “UUID", "revNr": "42“, "syncState": “UPDATED" } “tenantId": “1", • Separation of tenants with tenantId for each object • General data provided by JD • Sync has to respect tenants • Cost-efficient operation Cloud Backend DB: MySQL Tenant 1 Data Tenant 2 Data Tenant n Data … General Data iPad iPad iPad iPad iPad iPad iPad iPad iPad iPad iPad iPad Tenant 1 Tenant 2 Tenant n
21. © Fraunhofer IESE 21 Multiple Users Share one Device 7 iOS apps and devices not prepared for multi-user
22. © Fraunhofer IESE 22 Multiple Users Share one Device 7 • Login mechanism for users • Custom realization of SSO • Usage of keychain group access for SSO tokens • Sharing among apps via backend • Removing data of logged out users from current SQLite DB Device 1: iPad JobManager iOS App DB: SQLite CoreData Sync Engine UI User 1 Data User 2 Data
23. © Fraunhofer IESE 23 Connecting to the World 8 • Collecting data on the field • Streaming map data • Reusing weather data across tenants • Sharing master data with other management systems Machine Identification Machine Data Map Service Farm Management Weather Service Device 1: iPad iOS App Cloud Backend Logic DB: MySQL
24. © Fraunhofer IESE 24 Increasing Amount of Data 9 Cloud Backend DB: MySQL Hibernate Logic Mobile Services Device 1: iPad JobManager iOS App DB: SQLite CoreData Sync Engine UI Sync • > 10.000 objects • Continuously growing • Offline requires data sync • Slow initial sync (> 60s) • Filter old transactional data • Compress • Prefill SQLite DB file • Deeply optimize slow CoreData storage • …
25. © Fraunhofer IESE 25 Cloud Backend Increasing Frequency of Changes of Data 10 Near real-time data (GPS,…) → ~1 change/s Live distribution to devices Visualization and analysis • CQRS: Separate interface for high frequency data • Event-Sourcing • Cassandra for fast writes • In-memory for fast reads Device 1: iPhone Logistics iOS App DB: MySQL Hibernate Logic Mob. Serv. Sync HFR In Memory Cassandra HFW
26. © Fraunhofer IESE 26 Upgrading Apps 11 Device 1: iPad JobManager iOS App Device 2: iPad JobManager iOS App v1 v2 • Support multiple app versions • Maintain single DB model • Create mapper SQLite (model v1 ) SQLite (model v2 ) Cloud Backend DB: MySQL (model v2 ) Hibernate Logic S v1 + Mapper S v2
27. © Fraunhofer IESE 27 Multiple Languages to be Supported 12 Device 1: iPad JobManager iOS App Device 2: iPad JobManager iOS App English German UI UI Language resources for texts Translated master, reference, and transactional data Metrics / units • Where to transform units? • In which units to persist? • Tenant-spanning analyses Cloud Backend DB: MySQL Hibernate Logic Mobile Services Sync
28. © Fraunhofer IESE 28 Summary: Overview of Data Representations Cloud Backend DB: MySQL Hibernate Logic Mob. Serv. Sync HFR In Memory Cassandra HFW Device 1: iPad JobManager iOS App DB: SQLite CoreData Sync Engine UI Java Objects Cassandra Column F. Java DTOs Java Entities Relational Tables JSON DTOs Relational Tables CoreData Objects
29. © Fraunhofer IESE 29 Lessons Learned – Impact on Quality Attributes UX and performance: high impact on data architecture Maintainability: nearly always adversely impacted by concepts for other quality attributes Security: makes offline capability even more difficult
30. © Fraunhofer IESE 30 Lessons Learned – Technical Aspects Offline capability: not out of the box CoreData: requires in-depth technical know-how Offline capability: costly to develop
31. © Fraunhofer IESE 31 Lessons Learned – Data Modeling There is not THE single data model Data modeling according to OOD / DDD is not enough Data modeling is highly intertwined with architecture design
32. © Fraunhofer IESE 32 Key Takeaways Data is often neglected in architecture Data is highly related to quality attributes Design and document decisions around data! Farming is innovative and strongly uses IT ☺
33. © Fraunhofer IESE 33
34. © Fraunhofer IESE 34 Dr. Matthias Naab Phone: +49 631 6800 2249 matthias.naab@iese.fraunhofer.de Dr. Ralf Carbon Phone: +49 631 3619 1813 carbonralf@johndeere.com
35. © Fraunhofer IESE 35 Appendix: Legend of Diagrams Computational Node App Yellow: Introduced or changed here Component Data object Connector