GEAS:Generic Adaptive Scheduling For High-efficiency Context Inconsistency Detection

Context-aware applications can make smart adaptations by using contexts collected from sensors to understand their running environment.However,due to sensing noises,collected contexts can be inaccurate,incomplete,or even conflicting to each other,thus leading applications'abnormal adaptation or even failure.To address this problem,one promising approach is to check contexts against some pre-specified consistency constraints to detect context inconsistencies for applications.This process is known as constraint checking,and typically,the constraint checking process is scheduled every time collecting any context change.However,this naive scheduling strategy is low-efficient and almost impossible to deploy when it comes to heavy workload scenarios,even if applying some efficient incremental or parallel constraint checking techniques.One may break this low-efficiency problem by checking grouped context changes together,which is known as batch-based scheduling,but this can lead to severe inconsistency missing problem and thus causing applications'misbehaviors.Therefore,to address this dilemma,we propose a novel scheduling strategy GEAS,which can not only improve the context inconsistency detection efficiency,but also avoid missing any context inconsistency.The key insight behind GEAS is that when checking grouped context changes together,we observe that only grouping certain combinations of context changes together can cause missed context inconsistencies in the detection.GEAS smartly models such combinations that can cause missed context inconsistencies in inconsistency detection as susceptibility conditions(s-conditions),and then identifies and avoids grouped context changes matching s-conditions together at runtime.Moreover,what is also worthy noticing is that,since GEAS is a strategy for scheduling constraint checking,it actually complements existing efforts on efficient constraint checking in an orthogonal way,thus able to generically improve the efficiency of all existing constraint checking techniques.Specially,GEAS has three phases.First,GEAS statically derives s-conditions from concerned consistency constraints.Second,based on those statically derived s-conditions,GEAS at runtime identifies them when collecting context changes and proactively avoids grouping context changes that matches any s-condition.Finally,GEAS thus adaptively groups context changes together as long as no s-condition is matched among grouped changes.We also experimentally evaluated GEAS with four state-of-the-art constraint checking techniques on large-volume real-world taxi data,to validate the necessity and unique effectiveness of GEAS.The experimental results show that:(1)GEAS achieved 72%-539%efficiency improvement,as compared to the naive immediate strategy;(2)GEAS successfully avoided missing any context inconsistency in the detection while the traditional batch-based scheduling strategy caused 39.2%-65.3%missed context inconsistencies.Moreover,in order to further investigate GEAS'effectiveness under real-world heavy-workload scenarios,we also conduct a case study with practical settings.The case study results show that:(1)GEAS achieved 47%-446%efficiency improvement as compared to the immediate strategy;(2)GEAS also maintained a nearly zero inconsistency missing rate in the detection;(3)GEAS significantly enhanced the checking workload that can be handled by certain constraint checking techniques.