Hybrid Context Inconsistency Resolution for Context-aware Services Chenhua Chen1, Chunyang Ye2, 3 and Hans-Arno Jacobsen2 1Department of Computer Science, University of Saarland 2Middleware Systems Research Group, University of Toronto 3 Institute of Software, Chinese Academy of Sciences Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Outline Background Context-awareness Research Problem Context Inconsistency Resolution Hybrid Solution Context Correlation Model Application Recovery Model Experimental Results Chen, Ye and Jacobsen, PerCom'11, Seattle Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Context-awareness An important feature of pervasive applications Contexts locations, time etc. Implicit input/output Seamless integrated Context-awareness Sense environment automatically Remember history Adapt to changing situations Chen, Ye and Jacobsen, PerCom'11, Seattle Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Supply Chain Scenario Update warehouse database Reading RFID tags Chen, Ye and Jacobsen, PerCom'11, Seattle Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Context Inconsistency Reasons Environmental noise Examples RFID reader report wrong readings Register incorrect number in warehouse GPS or GSM devices report inaccurate location Pick wrong route Chen, Ye and Jacobsen, PerCom'11, Seattle Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Context Inconsistency Resolution Validate consistency constraints Consistency constraints Context queue Inconsistency resolution 2) Remove oldest 3) Remove all Inconsistent contexts 4) User preference, heuristics etc. 1) Remove latest Chen, Ye and Jacobsen, PerCom'11, Seattle Chenhua Chen, Chunyang Ye and Hans-Arno Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Limitations Difficult to identify problematic contexts E.g., remove the latest, oldest, least frequently used etc. Counter example to remove the latest Two RFID readers, the first one is inaccurate, the second one is accurate Resolution approaches rely heavily on constraints Accuracy and completeness of constraints are crucial Counter example Constraint: Two RFID readers report identical readings Reported readings are the same but inaccurate Chen, Ye and Jacobsen, PerCom'11, Seattle
Our Proposal: Hybrid Solution Chen, Ye and Jacobsen, PerCom'11, Seattle
Example of Our Proposal 1. Two readers report inconsistent readings 2. Postpone inconsistency resolution 3. Warehouse check in, collect weight info 4. Update profile of goods 5. Resolve inconsistent readings based on weight and profile Chen, Ye and Jacobsen, PerCom'11, Seattle
Challenges How to make use of the application semantics in resolution? Close to T0: Semantic information is of limited usefulness When to resolve? Close to T2: unacceptable recovery cost Chen, Ye and Jacobsen, PerCom'11, Seattle
Example of Application Semantics Previous Location: (2, 3) Current Location: (4, 5) Inconsistency found! The probability of each context being inaccurate is 50% Continue move one step New Location: (4, 4) warehouse (2, 3) is more likely to be inaccurate, since it is impossible to move from (2, 3) to (4, 4) in two steps. Chen, Ye and Jacobsen, PerCom'11, Seattle
Context-correlation Model CL NL C1 C2 C3 C4 C5 C7 C6 C8 C9 fe (c3, a) Current contexts Contexts after invoking action a fe(CL, a): | NL – CL|≤ 1 C3 C8 At least one of C3 and C8 is inaccurate! Chen, Ye and Jacobsen, PerCom'11, Seattle
Context-correlation Model … C7 C8 C9 Ci Cj Ck Context Ci1 Contexts Ci2 Contexts Cin C1 C2 C3 p1 p2 p3 p1 ≥ 1- p2 * p3
Application Error Recovery Inconsistency resolution s0 s2 a1 a2 s1 s3 a3 s4 a4 Sensing c detection s2’ s3’ b4 s2” b3 b2 Backward recovery Forward recovery Chen, Ye and Jacobsen, PerCom'11, Seattle
Example of Error Recovery Backward recovery Backtrack the movement Forward recovery Select a different path warehouse Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Cost Model Compensation cost (cpc) For backward recovery Cost of compensating a task Execution cost (ecc) For forward recovery Cost of executing a task Total cost for an error recovery plan Chen, Ye and Jacobsen, PerCom'11, Seattle
Resolution Algorithm Collect application semantics Inconsistency detected Postpone resolution Application continues Build correlation graph Compute error recovery cost Resolve inconsistency Calculate probability Error recovery Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Experiment Setup 16 X 16 Map cpc = ecc = 1 Search the target in a heuristic way Random placement of goods Metrics: Accuracy of resolution Cost of error recovery warehouse Chen, Ye and Jacobsen, PerCom'11, Seattle
Results L-RL: Remove latest L-RO: Remove oldest M-H: Hybrid solution Higher error rate Chen, Ye and Jacobsen, PerCom'11, Seattle
Results L-RL: Remove latest L-RO: Remove oldest M-H: Hybrid solution Location-aware Higher threshold Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Results L-RL: Remove latest L-RO: Remove oldest M-H: Hybrid solution H-ER: Error recovery only Higher error rate Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Results L-RL: Remove latest L-RO: Remove oldest M-H: Hybrid solution H-ER: Error recovery only Location-aware Higher threshold Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Scalability Randomly generate correlation graph Calculate probability of each context being inaccurate Record the time needed Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Conclusions A novel approach to resolve context inconsistency Combine low-level inconsistency resolution with high-level error recovery Correlation model to reason about inaccurate contexts Cost model to calculate recovery cost Algorithm to trade off accuracy against recovery cost Future work More real-life experiments Extend the correlation model to support confidence Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle
Chen, Ye and Jacobsen, PerCom'11, Seattle Related Work Existing resolution strategies [Heckmann, IJCAI-MRC’05] Remove the latest, the oldest, the least frequently used [Bu et al. QSIC’06] Remove all [Park et al. Compsac’05] User preference [Capra et al. TSE’03] Auction [Xu et al. ICDCS’08] Heuristics Chen, Ye and Jacobsen, PerCom'11, Seattle