Load Shedding in Mobile Systems With MobiQual

IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL. XX, NO. XX, XXX 2009 1 Load Shedding in Mobile Systems with MobiQual Bu ˘ gra Gedik, Member, IEEE, Kun-Lung Wu, Fellow, IEEE, Ling Liu, Senior Member, IEEE, Philip S. Yu, Fellow, IEEE Abstract — In loc ati on- bas ed, mob ile con tin ual que ry (CQ) sys tems, two ke y me asure s of quali ty of ser vic e (QoS) are: freshness and accuracy. To achieve freshness, the CQ server must perf orm fre quent quer y re- eval uatio ns. T o attai n accur acy , the CQ serv er must recei ve and proc ess freq uent position updat es from the mobile nodes. Howeve r , it is often difficu lt to obtai n fresh and accurate CQ results simultaneously, due to (a) limited resources in computing and communication and (b) fast-changing load condition s cause d by conti nuous mobile node movemen t. Hence, a key challenge for a mobile CQ system is: How do we achi eve the highest possibl e quality of the CQ resu lts, in both fre shnes s and accur acy , with curr entl y ava ilabl e res ourc es? In this paper, we formulate this problem as a load shedding one, and develop MobiQual − a QoS-a ware approac h to perf ormi ng both update load shedding and query load shedding. The design of Mo- biQual highlights three important features. (1) Differenti ated load shedding: We apply different amounts of query load shedding and update load shedding to different groups of queries and mobile nodes, respectively. (2) Per-que ry QoS specificati on: Individualize d QoS specific ation s are used to maximize the overall freshness and accuracy of the quer y results. (3) Low-cost adapt ation : Mob iQu al dyn ami cal ly adapts, wit h a mi nimal ove rhe ad, to changing load conditions and available resources. We conduct a set of comprehensive experiments to evaluate t he effectiveness of MobiQual. The results show that, through a ca reful combination of update and query load shedding, the MobiQual approach leads to much highe r fre shnes s and accurac y in the query results in all cases, comp ared to exis ting approach es that lack the QoS - awareness properties of MobiQual, as well as the solutions that perform query-only or update-only load shedding. Index T erms — Loc ation- bas ed sys tems, Query proc essin g, Load Shedding I. I NTRODUCTION Today, we are experiencing a world where we can stay con- nected while on-the-go because there are (1) myriads of affordable mobile devices and (2) ever increasing accessibility of wireless communications for these devices. Combined with the availability of lo w-c ost posit ioning dev ice s, lik e GPS , thi s has created a new class of appli catio ns in the area of mobile location- based servi ces (LBS s). Examp les inclu de locat ion- aware info rmat ion delivery and resource management, such as transportation services (NextBus bus locator [1], Google ride finder [2]), fleet management, mobile games, and battlefield coordination. A key challenge for LBSs is: How to design a scalable location moni tori ng system capable of handling a lar ge numbe r of mobile nodes and processing complex queries over their positions? Al- though several mobile continual query (CQ) systems have been proposed to handle long-running location monitoring tasks in a • B. Gedik and K-L. Wu are with the IBM T.J. Watson Research Center, 19 Skyline Dr., Hawthorne, NY 10532. E-mail: {bgedik,klwu,psyu}@us.ibm.com. • L. Liu is with the College of Computing, Georgia Institute of Technology, 801 Atlantic Drive, Atlanta, GA 30332. E-mail: [email protected]. • P. S. Yu is with the Computer Science Department, University of Illinois, 851 S. Morgan Street, Chicago, IL 60607. E-mail: [email protected]. scalable manner [3], [4], [5], [6], [7], [8], [9], [10], [11], [12], [13], the focus of these works is primarily on efficient indexing and query processing techniques, not on the accuracy or freshness of the query results. Acc ura cy (in acc ura cy) is mea sur ed bas ed on the amo unt of mobile node position errors found in the query results at the time of query re-eva luat ion. This accur acy measu re is stron gly tied to the freq uency of posit ion updates receiv ed from the mobil e nodes. Al though one can al so use a hi gher le ve l concept to measure accuracy, such as the amount of containment errors found in the que ry res ult s 1 , incl uding both fal se posit ive s (incl usion errors) and false negatives (exclusion errors), we argue that using position update errors for accuracy measure will provide a higher le ve l of pre cis ion . Thi s is pri mar il y becaus e by uti liz ing the amo unt of node pos iti on err ors as the accur acy measu re, one can easily bound the inaccuracy by a thre shold -based position repor ting scheme [3], [14]. Note that certain appli catio ns hav e higher tolerance to inaccuracy in position updates, such as region- based traffic density monitoring; whereas certain others require higher accuracy, such as path-based location tracking. Freshness (staleness), on the other hand, refers to the age of the query results since the last query re-evaluation. It is dependent on the frequency of query re-evaluations performed at the server. As mobile nodes continue to move, there are further deviations in mobile node positions af ter the last query re-evaluation. However, such post-query-re-e val uat ion dev iat ion s are not att ri but ed to inaccuracy . Hence, freshness can be seen as a metric capturing the se de via tio ns. It is impo rt ant to note tha t hig her fr eshnes s does not necessarily imply higher accuracy and vice versa. The concepts of freshness and accuracy in mobile CQ systems are, to some extent, similar to those of timeliness and completeness, respe cti vely , in the web infor mati on moni tori ng domai n [15]. Note that certain applications have higher tolerance to staleness in query results, such as monit orin g slo wly changing temporal aggregates like traffic rates; whereas certain others require higher freshness, such as location based alerts and triggers. Accuracy and freshness are not completely independent of each other in the sense that both reflect the uncertainty about the precise locat ions of the moving objects at a giv en time . Howeve r, we use accuracy and freshness separately to capture the impact of such uncer tain ty from two indep endent persp ecti ves. Accur acy is use d to captur e the effe ct of unc ert ain ty due to loc ati on- update frequency on the mobile devices, while freshness is used to capture the effect of uncer taint y due to query re-ev aluation frequency on the server. To obtain fresher query results, the CQ server must re-evaluate the continual queries more frequently, requiring more computing resou rces. Similar ly , to attai n more accurate quer y resul ts, the CQ server must recei ve and proce ss posit ion updates from the 1 A bound on the amount of containment errors can be approximated by a bound on the position errors, if the distribution of the mobile nodes around the query boundaries is at hand or can be approximated. IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING, VOL.23, NO. 2, Feb. 2011 h t t p : / / w w w . i e e e x p l o r e p r o j e c t s . b l o g s p o t . c o m

Load Shedding in Mobile Systems With MobiQual

Documents

Transcript of Load Shedding in Mobile Systems With MobiQual