This document presents a framework for modeling and analyzing the performance of context-aware mobile software systems. The framework uses statecharts to model context evolution across different dimensions like physical location, logical location, and hardware configuration. It allows specifying alternative system behaviors based on "if context is" conditions. Context-specific annotations are added to behavior models to estimate resource usage. Different analysis models can then be derived and solved to obtain performance indices for specific contexts. The document demonstrates the framework on an e-health system example and compares response times for two mobility scenarios. Future work includes modeling more complex context compositions and adaptation at runtime.

1. Fundamental Approaches to Software Engineering (FASE 2010, Paphos, Cyprus) Performance Modeling & AnalysisofContext-aware Mobile Software Systems

2. motivation Pervasiveness of software systems running on resource constrained, portable devices Providing the best user experience despite limited and variable amount/types of hw resources physical mobility of nomadic users logical mobility of software components Need to analyze context-aware software systems from a non-functional viewpoint 24/03/2010 Performance Analysis of Mobile Context-Aware Sw Systems 2 Providing methodologies, techniques and tools to develop and analyze software systems that need to be adaptable to provide the best user experience.

3. Introduction: Context, Mobility and Context-Awareness Our framework: Awareness Managers, Adaptation driven by “if context is” conditions, Context Manager Performance Analysis Conclusions and Future Work 24/03/2010 Performance Analysis of Mobile Context-Aware Sw Systems 3 Outline

5. Logical Locationof software components(i.e. the current deployment of sw components on execution hosts)

7. Logical Mobility: software components are redeployed on different execution hosts

9. Logical Location Awareness: capability to “sense” the current Logical Location

11. It can be restricted for those cases where only few context attributes are necessary (e.g. logical mobility not allowed)Without modifying the framework structure

12. UML-based Modeling of Users and Services Software Architecture, Behaviors, Hardware Platform Performance parameters 24/03/2010 Performance Analysis of Mobile Context-Aware Sw Systems 6 The framework Mobile Mobile Adaptable Configurable Context-specific Context, a cross cutting concern

13. A leading example: E-health system The eHealth System supports the doctor’s everyday activities providing distributed services, such as the retrieval of information about patients. The doctor invokes the service using a resource-constrained PDA. He can move across different locations: his home, the surgery, the patient’s home. The system performance can be affected by the doctor’s physical location and by the hardware configuration of the doctor’s PDA. The system performance is affected by the context 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 7

14. E-Health: UML design Model at a glance Software Architecture (Component Diagram) Services (Use Case Diagram) Service Behavior (Sequence Diagram) hw configurations Static <<deployment>> The doctor can display the medical history of a patients (text and images) Hardware Platform (Deployment Diagram) 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 8

16. Is a stochastic statechart where

17. A state represents the (set of) attribute(s) value(s) of the context-related model element

18. Transitions represent the events triggering changes in attribute values

19. Probabilities are associated to transitions

20. Is able to interact with other managers through remote firing

21. Defines a Variable that indicates its current state (e.g. $manager = ‘A’ , $manager = ‘B’)3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 9 MANAGER 1-p B A p

22. E-Health: context modeling Services (Use Case Diagram) Component Diagram Client :: LOGICAL MOBILITY Doctor :: PHYSICAL MOBILITY BATTERY :: HW CONFIGURATION Binding between hardware resources and physical places Binding between software resources and hardware platform Hw platform CPU:: HW CONFIG Physical Places remote firing Dynamic Deployment Diagram home surgery Performance Analysis of Mobile Context-Aware Sw Systems 10

23. The framework : behavior adaptation Adaptation: capability to change software behavior w.r.t. changes in the sensed context. We enable adaptation by allowing the system to choose among different implementations jof the same service i 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 11 Service i Start (i.e. service i invocation) “If context is” condition “If context is” condition “If context is” condition Si,1 Si,j Si,m Given the context description, how to represent “if context is” conditions?

24. E-Health: behavior adaptation (1) “If context is” condition Start RequestPatientInfoPage DISPLAY::$HwConfig==‘Color‘ AND DOCTOR::$PhyLoc==‘Surgery’ cond cond “If context is” condition IMG NO IMG DISPLAY::$HwConfig==‘B/W‘ AND DOCTOR::$PhyLoc==‘OpenAir’ Behavior Choice for each Service We specify “if context is” conditions using manager variables in logical predicates to relate alternative behavior descriptions to the modeled context 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 12

26. estimated software resource demand for each message 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 13

27. E-Health: context-specific annotations Start RequestPatientInfoPage IMG NO IMG <<GaAcqStep>> {acqRes = Instr, Msg, DbAx resUnit=“(5,2,2)”} <<GaAcqStep>> {acqRes = Instr, Msg, DbAx resUnit=“(5,1,2)”} <<GaWorkloadEvent>> {pattern=closed, population=200, extDelay=(0.5, sec)} Behavior Choice for each Service (Interaction Overview Diagram) Resource Constrained Behavior The doctor doesn’t dowload images THEN There is a different resource usage 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 14

29. is associated to the whole system (model)

30. is a stochastic statechart where

31. A (super) State models the combination of awareness manager states

33. CONTEXTMANAGER E-health: the context manager Doctor :: PHYSICAL MOBILITY false true DISPLAY :: HW CONFIG Each superState represents a context for service provision SR = @Surgery the Resource Constrained service implementation is invoked, i.e. NO IMAGES 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 16

34. E-health: the context manager GOAL Performance Analysis: Response Time in 2 different scenarios: i) Basic Scenario, ii) High Physical Mobility Scenario The Scenarios differ for transition probabilities of Physical Mobility Manager Basic High Mobility DOCTOR::PHYSICAL MOBILITY DOCTOR::PHYSICAL MOBILITY DISPLAY::HW CONFIGURATION 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 17

35. E-health: the context manager Basic and High Mobilityscenarios induce the samesuperstates and supertransitionsbutdifferenttransitionprobabilitiesthatleadtodifferentsteady state probabilties Basic High Mobility CONTEXT MANAGER (Steady State Prob) CONTEXT MANAGER (Steady State Prob) @SURGERY = 0.85 @OPEN AIR= 0.1216 @OPEN AIR = 0.4 @every other place = 0.2 Foreachsuperstatewe generate and solve a different performance model and weigh the resultingindiceswith the steady state probabilities 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 18

37. Petri Nets

38. Execution Graphs (no resource contention)Steady State Probability SW resourcedemands <<nfp>>ServiceTime : NFP_DataTxRate =1bit/μs x sw2hw conversionfactors UML Model + MARTE SDs▶ DD ▶ = HW resourcedemands x Service Timeofhwresources (e.g. CPU,DISK) services = Context–specific performance indices(e.g. ResponseTime) 24/03/2010 Performance Analysisof Mobile Context-AwareSwSystems 19

39. E-health: performance analysis MAX ResponseTime 82.07 sec CONTEXTMANAGER Average: Basic : 14.59 sec High Mobility: 26.32 sec MIN ResponseTime 1.17 sec Lower Average Response Time in Basic Scenario due to different Doctor’s Physical Mobility: more time spent @Surgery with faster internet connection 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 20

40. conclusions and future work We introduced a framework for modeling and analyzing performance of context-aware mobile systems Context-awareness is a composite/cross-cutting concern: context dimensions can be added/removed/ignored using stochastic statecharts and “if context is” conditions UML-based realization: extensible with (standard) profiles to other UML-based model-driven NFPs analyses [Future]: we are working on complex composition of managers [Future]: Modeling Adaptation@runtime where context state changes during service execution. 3/24/2010 Performance Analysis of Mobile Context-Aware Sw Systems 21