Embed Size (px)
Citation preview
Computers in Human Behavior 30 (2014) 485–490
Contents lists available at SciVerse ScienceDirect
Computers in Human Behavior
journal homepage: www.elsevier .com/locate /comphumbeh
Automated and user involved data synchronization in collaborativee-health environments
0747-5632/$ - see front matter � 2013 Elsevier Ltd. All rights reserved.http://dx.doi.org/10.1016/j.chb.2013.06.019
⇑ Corresponding author. Tel.: +966 14676189.E-mail addresses: [email protected] (M.S. Hossain), [email protected] (M.
Masud), [email protected] (G. Muhammad), [email protected] (M. Raw-ashdeh), [email protected] (M. Mehedi Hassan).
M. Shamim Hossain a,⇑, Mehedi Masud b, Ghulam Muhammad a, Majdi Rawashdeh c,Mohammad Mehedi Hassan a
a College of Computer and Information Sciences, King Saud University, Saudi Arabiab Computer Science Department, Taif University, Saudi Arabiac EECS, University of Ottawa, Ottawa, Canada
a r t i c l e i n f o a b s t r a c t
Article history:Available online 13 July 2013
Keywords:e-HealthCollaborative environmentData synchronizationUser interaction
This paper presents a data synchronization model using automated and user involved process during exe-cution of conflicting updates. Data synchronization is performed using three techniques, namely, (i) autosynchronization, (ii) semi-automatic synchronization, and (iii) user-involved synchronization. We haveevaluated and measured users’ acceptability of the proposed data synchronization approach in ane-health environment. The results show the effectiveness of the proposed approach.
� 2013 Elsevier Ltd. All rights reserved.
1. Introduction
A collaborative e-health system provides an environment toenable collaborative treatments through sharing of data amongthe healthcare service providers (e.g., physicians, hospitals, labora-tories, etc.). In such an environment service providers are autono-mous, therefore, independently curate, revise, and extend theshared data. For full collaboration the service providers exchangedata updates. Hence, data updates (insert, delete, or modify) in aservice provider, i.e., data source may, in turn, affect the data inthe other service providers.
Consider an e-health system, where family physicians, walk-inclinics, hospitals, medical laboratories, pharmacists, and otherstakeholders are willing to share information about patients’ treat-ments, medications, and test results. These sources need to coordi-nate their data of patients. Coordination may mean something assimple as propagating updates to each other. For example, a familyphysician and a hospital may want to coordinate the informationabout a particular patient. A hospital may store data about thepatient’s special treatment results while the patient was in thehospital. The family physician stores information about patient’sregular treatments data. The family physician which is not storingspecial treatments data, would benefit by the exchange of updatesmade on the hospital data source. On the other hand, the hospital
data source would benefit by the exchange of updates of any pre-vious diagnosis results related to the patient in the family physi-cian data source.
Exchanging data through update propagation in a collaborativee-health system is different from update processing in a traditionalreplicated database system (Masud, Kiringa, & Ural, 2009). First, areplicated database system assumes that the same data isreplicated in different data sources to increase performance andavailability of data. Meanwhile, the data stored in the sources ina collaborative e-health system is not replicated and may usedifferent domains (Kementsietsidis et al., 2003). Second, an updateoriginated in a source in a replicated system must be executed atall the sources to maintain consistency and ensure a single logicalview of data throughout the network (Masud et al., 2009). Incontrast, in a collaborative e-health system, an update may notneed to execute at all the service providers. Rather, it is sufficientto execute the update only at the relevant sources to the update.
1.1. Contribution
This paper presents an approach for data synchronizationwhere sources resolve conflicts in a collaborative fashion. Mainlydata synchronization is performed using three approaches:
Auto Synchronization (Auto-Sync): In auto-sync, sourcesdetermine the execution order of conflicting updates for data syn-chronization through collaboration. In auto-sync, no user involve-ment is necessary to order conflicting updates. Auto-Syncsynchronization is performed using the absorption resolution rule(Santoro, 2006).
486 M.S. Hossain et al. / Computers in Human Behavior 30 (2014) 485–490
Semi Automatic Synchronization (SemiAuto-Sync): In thisapproach, data synchronization is performed through the collabo-ration of sources and with users’ involvement. The SemiAuto-Syncapproach is applied when two conflicting updates are executedinto the same number of sources and the Auto-Sync approach failsto resolve the conflict. This synchronization is performed usingmutual resolution rule.
User-involved Synchronization (User-Sync): In this approachusers are directly involved to resolve a conflict of updates for datasynchronization. Mainly, the conflicts are resolved by the superusers. User-Sync approach is used in some critical situations whereAuto-Sync process cannot reach a decision for resolving conflictsamong updates.
In general, our proposed approach is applicable where systemstolerate data inconsistency for a certain period of time and updateto a data in a source is not immediately important to the othersources. However, eventual consistency (Saito & Shapiro, 2005;Shapiro & Kemme, 2009) is guaranteed.
The next section presents notions of ensuring consistency andthe approach of data synchronization. The notions are used to pres-ent the constraints for maintaining data consistency. Section 3 de-scribes evaluation results. Section 4 describes the related works.Finally, Section 5 concludes the work remarking and pointing outavenues for further research.
2. Data synchronization
In a distributed e-health system updates are executed locallyand independently. The system does not require a multi-sitecommit protocol (Hwang, Srivastava, & Li, 1994) (e.g., two phasecommit), which leads to introduce blocking and is thus not feasi-ble. Specifically, updates are executed locally, and then asynchro-nously propagated over acquainted sources. A consistentexecution of updates can be obtained by ensuring the same execu-tion order of the conflicting updates over each acquainted source inthe propagation path of the updates. Authors in Masud and Kiringa(2011) describe the notion of consistent execution of conflictingupdates for data synchronization. In the following, we proposean approach to achieve consistent execution of conflicting updatesin automatic and user-involved process.
In order to maintain data consistency in the acquainted sourcesthe execution order of the conflicting updates must be same in allthe sources. If there is no conflict at the time updates are initiatedin a source, then the updates can be executed in any order in theacquainted sources. Therefore, when two updates D1 and D2 areexecuted at a source and are not conflicting updates, then their dif-ferent execution order in the acquainted sources of the source doesnot create any inconsistency.
An optimistic approach is used for executing updates in the datasources. Optimistic approaches allow continuous data access dur-ing update execution. They allow users to read or update the data-base while they are disconnected and synchronize the data withother sources when they reconnect (Kermarrec, Rowstron, & Shap-iro, 2001; Petersen, Spreitzer, Terry, & Theimer, 1997; Terry, Thei-mer, Petersen, Demers, & Spreitzer, 1995). Here updates arepropagated asynchronously in the background without blockingany read requests. Many major commercial database vendors sup-port this mode of asynchronous replication. The steps to resolveconflict and synchronizing updates are given below.
Step 1: When a source detects a conflict between two updatesthat are received from other sources then the source stops exe-cution of the updates and asks its parent about the executionorder. Note that a source which forwards an update is a parentand the source which receives the update is called a child.
Step 2: If the parent has no information of execution, then theparent asks its parent. This process continues until a source isfound that knows the order or the inquiry reaches to the updateinitiators. The first source which detected the conflict of thesame pair of updates may have already propagated an inquiryto the initiators and the execution order is already decided.Hence, other sources which detect the same conflict mayreceive the result from any intermediate sources along the pathto the initiator.
Now, we describe the process of selecting the execution order oftwo conflicting updates by the initiators or by an intermediatesource who has the order information. An order is selected usingthree resolution protocols. The protocols are mutual resolution,absorption resolution and user-involved resolution. The mutual reso-lution is an automated process, the absorption resolution is a semi-automated process, and the third one is user-involved process.
Mutual Resolution: From the conflict message each initiatorknows how many sources have executed the updates since theconflict message has travelled through a path from the sourceswhere the conflict is detected to the initiators. We treat this countas execution level of an update. The execution level of an update Di
is denoted as level(Di). If (level(D1) = level(D2)) then the order isdetermined with mutual understanding. After receiving theconflict information, initiators agree on an specific order. Considerthe order hD2,D1i is selected. After selecting the order the follow-ing process starts.
A compensate update D�1 is generated and is sent to Si. When Si
receives the order information and the compensate update, Si exe-cute D�1 and execute the updates D1 and D2 in the order hD2,D1i.
Absorption Resolution: After receiving the conflict information,both the initiators agree on an order hD2,D1i if level (D2)>level(D1)else the order is hD2,D1i. The compensation process starts as de-scribed in mutual resolution.
User-involved Resolution: If a conflict is detected at source Si andSj, the super users in the sources decide the update executionorder. The users are informed about a conflict when a conflict isdetected by the system. If the users cannot decide the order thenthe users consult with the users of the parents of the updates’propagator. After the conflict is resolved the compensation processstarts as described in mutual resolution.
3. Evaluation
For evaluating the proposed data synchronization approach wemeasure the feasibility, efficiency, and users’ acceptability of thesystem. Each data source is populated with a few hundred differentrelational tuples. In order to generate data, we wrote a simulatorusing Java. We choose integers value for the domain of the attri-butes in each relation. Acquaintances are built based on data itemsstored in each source. Each update is either insertion, deletion, orreplacement of some tuple in a source. We consider SQL for updateoperations. All experiments were performed at least five times andwe took the average of these measurements.
3.1. System evaluation
We evaluated the proposed update synchronization approachconsidering different conflict factors of updates. A conflict factordenotes the ratio of the number of updates that are involved inconflict and the total number of updates that are active in the sys-tem. For example, 0.2 conflict factor means that 20% of the totalupdates that are active in the system are involved in conflict. Notethat updates are generated from different sources. The objective ofthis evaluation is to examine what is the effect of the conflict
Fig. 1. Execution time for the number of conflicting updates.
M.S. Hossain et al. / Computers in Human Behavior 30 (2014) 485–490 487
factors to the proposed update conflict resolution mechanism. Weperformed two experiments with two different settings to examinethe effect.
In the first experiment, we fixed the number of sources totwenty but varied the number of updates generated from thesources. Fig. 1 shows the results of the experiment. Note that weselected the updates in such a way that updates involve in conflictaccording to the specified conflict factors (0.1, 0.2, 0.3, 0.4, and 0.5).From the result, we observe that the execution time increases withincreased value of conflict factors, however, the effect on executiontime is not a major inhibiting factor. Consider the execution time of20 updates with the conflict factors 0.4 and 0.5. We see that theexecution time are 64.264 and 67.82, respectively. Notice thatthe execution time increases slowly with the increased conflict fac-tors. We also observe that the execution time increases rapidlywith increased number of concurrent updates.
For the second experiment, we fixed the number of concurrentupdates generated from varied number of sources and conflict fac-tor in the system. The objective of this experiment is to examinethe behavior of the update propagation taking into account the dif-ferent number of sources with different conflict factors. Fig. 2shows the results of the experiment. We observe from the experi-ment that major time changes occur as a function of number ofsources in the system.
3.2. User acceptability evaluation
In this experiment, we measure user acceptability for the pro-posed data synchronization model in an e-health environmentadopting the Ubiquitous Computing Acceptance Model (UC-AM)(Hossain, Nazari, Alghamdi, & El Saddik, 2012; Spiekermann,2008). Spiekermann (2008) conducted an empirical research tofind out the impact of control on ubiquitous pro-active system.
Fig. 2. Execution time for the number of conflicting updates in different size ofnetworks with different conflicting factors.
Hence, it will be relevant to use UC-AM to analyze the level ofacceptance (as a measure of user acceptability) for level of userinvolvement in terms of users’ acceptability, consistence, correct-ness and risk. Using the UC-AM evaluation model, Spiekermann at-tempted to capture user acceptance in terms of several parametersrelated to their interest. Hossain et al. (2012) have chosen a se-lected set of parameters from the UC-AM model that are more re-lated in their context of interaction model, which are (1) perceivedconsistency, (2) perceived acceptability, (3) perceived correctnessand (4) perceived risk. Similar to both of the approaches, we alsohave chosen the same four parameters. According to the UC-AMmodel and user interaction model, our goal is to investigatewhether the proposed mechanism has positive impact on itsacceptability in e-health environment settings. In Table 1, we pro-vide the questionnaires for collecting users’ feedback after theexperiment was conducted. These questionnaires are adapted fromSpiekermann (2008) and Hossain et al. (2012) with some modifica-tions to suit our needs related to our context.
3.2.1. SettingFor the experiment, we considered 30 users for evaluating the
impact of user acceptability on the data synchronization. We di-vided the users into two groups, namely General Users and TrustedUsers, each group consisting of 15 users, respectively. The trustedusers are the initiators of the updates or can be trusted for decidingthe update order. They also have knowledge of the data to be up-dated by the update transactions, while General Users do not havesuch knowledge.
At first, participants of General Users and Trusted Users werebriefed about the synchronization: Auto-Sync, SemiAuto-Sync,and User-Sync so that they know what to expect from the system.They were explained for how the system works and when they willbe involved in data synchronization phase. For example, if a con-flict occurs then the users know how to resolve the conflict andwhat is the nature of the conflict. In the following section, we ana-lyze the results in terms of user acceptance.
3.2.2. User acceptance measureAs users’ experience is important for the synchronization of up-
dates, we have to understand the characteristics and needs of thereal target end users.
We summarize users’ response from the questionnaire of Ta-ble 1 related to data synchronization in order to better understandtheir perception on Auto-Sync, SemiAuto-Sync, and User-Sync inorder to qualitatively analyze the results. Further, 80% of the par-ticipants in Trusted User and 24% of the participants in GeneralUser had used system extensively, while 11% of the participantsof Trusted User and 76% of the participants of General User hadused the system only a few times. Fig. 3 shows their responses inpercentage scale.
We also notice that a significant number of users are not happywith Auto-Sync in terms of perceived acceptability, perceived cor-rectness and perceived risk. However, they mostly agree that thesynchronization is effective if user is involved during update syn-chronization. Similar responses have found for users with Semi-Auto-Sync, however, with their responses in low percentage. Onthe other hand, for User-Sync, the users were more optimistic.
As shown in Fig. 4, we also notice the responses for both groupof the users (1) Trusted User, and (2) General User for in threetypes of data synchronization: Auto-Sync, SemiAuto-Sync, andUser-Sync. It is noticeable that both the user groups respondedquite similarly with respect to the different measurement units.This may be due to the limited number of users and theirexperience. We however observe that all these users providedlower score in terms of perceived acceptability and perceived cor-rectness on Auto-Sync as shown in Fig. 4(a). For User-involved
Table 1Questionnaires used for update synchronization with regards to users’ acceptance.
No. Measurementvariable
Questionnaire
1 Perceivedconsistency
hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i would provide consistent health-care data for a certain period of timeI find hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i maintaining data consistency among the e-health service providers
2 Perceivedacceptability
I find hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i suitable for synchronization of conflicting updatesI find hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i providing relaxed consistency guaranteesAcceptable synchronization of conflicting updates can be obtained from hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i in a collaborativefashion
3 Perceivedcorrectness
I think that hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i allow continuous correct or reliable data access during update executionThrough the usage of hAuto-Sync.jUser-Sync.jSemiAuto-Sync.i I think the system can provide effective evaluation results
4 Perceived risk I can predict that in emergency healthcare situation, the system can resolve conflicts in a collaborative fashion without much humanintervention
Fig. 3. Comparison of user responses for the evaluation.
488 M.S. Hossain et al. / Computers in Human Behavior 30 (2014) 485–490
Synchronization (User-Sync), we summarize the group observationin Fig. 4(b). We also found similar score compared to Auto-Sync interms of perceived acceptability and perceived correctness onSemiAuto-Sync as shown in Fig. 4(c). We have found that both usergroups have higher perception of consistency, acceptability andcorrectness.
For comparing among the synchronization updates with respectto the different measurement variables, we present Fig. 5 that in-cludes the mean responses of the respective system’s users irre-spective of any user group. We observe that the update withUser-Sync have higher perceived consistency, acceptability andcontrol than Auto-Sync, which provide confidence to users’evaluation.
Fig. 4. Comparison of mean responses between two user groups for Auto
The overall scores provided by the users for the systems aresummarized in Table 2. The lower mean value of perceivedacceptability and control for Auto-Sync reflects poor acceptanceby the user. The moderate mean value of perceived consistencyshows that Auto-Sync has the potential even if the users pro-vided lower score in other variables. This is because users wantsome automation that can provide some benefit to them, butthey are just worried that the automation performed might berisky and they will have less control over the system’s function-ality. We also present a similar analysis in the case of User-Sync.Here, the higher mean value of perceived consistency, accept-ability and control represents a strong acceptance of the adap-tive interaction model. Also, note that users perceive less riskin User-Sync than in Auto-Sync, although both the system isattributed as moderately risky, which can be minimized withimproved design.
We have also performed ttest on the responses received fromTrusted Users (M = 2.60) and General User with 95% confidenceinterval and with a p-value less than 0.1. No significant differenceswere found between Trusted Users (M = 3.20, SD = 0.43733) andGeneral User (M = 2.73, SD = 0.24890) for participant ratings con-cerning perceived correctness t(28) = 0.582, p = 0.582. Significantdifferences were found between Trusted Users (M = 3.257) andGeneral User (M = 2.588) in terms of participants ratings concern-ing perceived consistency on e-health data with 95% confidenceinterval, two tailed p � 0.0001. For this test of perceived consis-tency, the intermediate values used in calculations are t = 7.9847,df = 14 and standard error of difference = 0.084.
-Sync in (a), User-Sync in (b), and SemiAuto-Sync in (c), respectively.
Fig. 5. Comparison of mean responses irrespective of user groups for Auto-Sync, SemiAuto-Sync, and User-Sync shows that the proposed interaction mechanism in User-Syncprovides better consistency, acceptability and correct update result with minimized risk than Auto-Sync, and SemiAuto-Sync.
Table 2User Acceptance based on the Overall Evaluation.
DataSyncronization
Means Perceivedconsistency
Perceivedacceptability
Perceivedcorrectness
Perceivedrisk
Auto-Sync. Mean 3.63 2.83 1.86 4.22Variance 0.37 0.47 0.31 0.41
User-Sync. Mean 4.00 4.20 4.08 2.90Variance 0.52 0.25 0.50 1.16
Semi-auto Mean 4.00 4.20 4.08 2.90Variance 0.52 0.25 0.50 1.16
M.S. Hossain et al. / Computers in Human Behavior 30 (2014) 485–490 489
4. Related works
Users involvement is important in assessing e-health domain tofind the correctness, acceptability and inconsistency. Some of thenotable research (Annie Jin & disclose, 2012; Behm-Morawitz,2013; Deed & Edwards, 2011; Forment, De Pedro, Casa, Piguillem,& Galanis, 2010; Hossain, 2011; Weaver, Thompsonb, Weavera, &Hopkins, 2009; Welbourne, Blanchard, & Wadsworth, 2013) in thisaspects are as follows.
Welbourne et al. (2013) discussed about Connectedness, Per-ceived stress, and Control variable for virtual health communities.In the paper, connectedness and control are used to measure thedegree of emotional feeling and frequency of reading message,while in our case Connectedness refers to the correct and reliablehealth data access. Behm-Morawitz (2013) presents a work relatedto an empirical study of the role of self-presence in a social virtualworld on individuals offline health, appearance, and well-being.Annie Jin and disclose (2012) describe users evaluation of privacymanagement in e-health and practical implications for the designof persuasive e-health websites are discussed.
Mork et al. (2004) proposed a framework for managing updatesin a large-scale data sharing system with the concept of view main-tenance. In the system, a peer which acts as a data receiver, itsschema is defined as a view of the schema of data providers.
Bertossi and Bravo (2007, 2008) propose a semantics for main-taining data consistency in peer data exchange systems. The incon-sistency between databases of peers is managed at query timeusing the repair semantics (Arenas, Bertossi, & Chomicki, 1999).However, in this paper, we investigate an update exchange mech-anism in data sharing systems where each peer typically contrib-utes its own data and may import data from other related peers.
Taylor and Ives (2006) proposed a database reconciliationmechanism in a decentralised collaborative data sharing environ-ment. Here conflicts are resolved using the priority of updatesand the provenance information. The approach requires centralisedprovenance information for resolving conflicts. Otherwise, sameupdate may be accepted by one peer and rejected by another peer.
Terry, Theimer, Petersen, Demers, and Spreitzer (1995) pro-posed a replicated database system to support collaborationamong users in a weakly connected network. Transactions arebroadcast between sites using an epidemic propagation protocol.It first executes transaction in their tentative order, then rollsback and replays them in final order. If the transaction is ac-cepted by a primary site, the final timestamp is assigned tothe update. Hence, the final execution of transactions relies ona primary site that enforces a global continuous order on agrowing prefix of history.
5. Conclusion
This paper presents mechanisms to support data synchroniza-tion through update exchange among the e-health service provid-ers. Updates are synchronized automatically by the sources using acollaborative fashion through exchange of conflict information. Insome cases users are involved to decide update order for data syn-chronization. Data synchronization is performed using three ap-proaches, namely, (i) auto synchronization, (ii) SemiAutosynchronization, and (iii) user-involved synchronization. In autosynchronization the sources resolve conflict between updatesusing absorption resolution rule. The absorption rule gives higherpriority to the update which has the higher execution level. There-fore, initiators can resolve a conflict automatically. Meanwhile, inSemiAuto synchronization, user’s involvement is necessary in or-der to decide the update order. This approach is used when con-flicting updates have same execution level. However, sourcesresolve conflict automatically if any intermediate source knowsthe update execution order information. In user-involved synchro-nization, super users mainly decide the update order. We imple-mented the approaches and evaluated the system performancesconsidering conflict factors and users acceptability. The evaluationresults show the effectiveness of the proposed approaches.
Our future goal is to extend the proposed approach consideringthe provenance information and the trust relationship among the
490 M.S. Hossain et al. / Computers in Human Behavior 30 (2014) 485–490
sources. The provenance and trust relationship will give a general-ized update model for data synchronization in e-health systems.
Acknowledgments
The authors extend their appreciation to the Deanship of Scien-tific Research at King Saud University for funding this workthrough the research group Project No. RGP-VPP-228.
References
Annie Jin, S. (2012). To disclose or not to disclose, that is the question: A structuralequation modeling approach to communication privacy management in e-health. Computers in Human Behavior, 28(1), 6977.
Arenas, M., Bertossi, L. E., & Chomicki, J. (1999). Consistent query answers ininconsistent databases. In Proc. of the ACM symposium on principles of distributedcomputing.
Behm-Morawitz, E. (2013). Mirrored selves: The influence of self-presence in avirtual world on health, appearance, and well-being. Computers in HumanBehavior, 29(1), 119128.
Bertossi, L. E., & Bravo. L. (2007). The semantics of consistency and trust in peer dataexchange systems. In Proc. of the Intl Conf. on logic for programming artificialintelligence and reasoning (pp. 107–122).
Bertossi, L. E., & Bravo. L. (2008). Information sharing agents in a peer data exchangesystem. In Proc. of the Intl Conf. on data management in grid and P2P systems(Globe).
Deed, C., & Edwards, A. (2011). The role of outside affordances in developingexpertise in online collaborative learning. International Journal of KnowledgeSociety Research, 2(2), 25–36.
Forment, M. A., De Pedro, X., Casa, M. J., Piguillem, J., & Galanis, N. (2010).Requirements for successful Wikis in collaborative educational scenarios.International Journal Knowledge Society Research, 1(3), 44–58.
Hossain, M. S. (2011). Adaptive media service framework for health monitoring. InProc. ACM ICIMCS’11. Chengdu, China, August 5–7, 2011.
Hossain, M. A., Nazari, A., Alghamdi, A., & El Saddik, A. (2012). Adaptive interactionsupport in ambient-aware environments based on quality of contextinformation. Multimedia Tools and Applications.
Hwang, S., Srivastava, J., & Li, J. (1994). Transaction recovery in federatedautonomous database systems. Distributed and Parallel Databases, 2(2),151–182.
Kementsietsidis, A. Arenas, M., & Miller. R. J. (2003) Mapping data in peer-to-peersystems: semantics and algorithmic issues. In SIGMOD.
Kermarrec, A. M., Rowstron, A., Shapiro, M., & Druschel. P. (2001). IceCube approachto the reconciliation of divergent replicas. In Proc. of the ACM symposium onPrinciples of Distributed Computing (PODC) (pp. 210–218).
Masud, M., & Kiringa, I. (2011). Transaction processing in a peer to peer databasenetwork. Data Knowledge Engineering, 70(4), 307–334.
Masud, M., Kiringa, I., & Ural, H. (2009). Update processing in instance-mapped P2pdata sharing systems. International Journal of Cooperative Information Systems,18(3), 339–379.
Mork, P., Gribble S. D., & Halevy. A. Y. (2004). Managing change in large-scale datasharing systems. In UW CSE Technical Reports UW-CSE-04-04-01.pdf.
Petersen, K., Spreitzer, M., Terry, D., Theimer, M., & Demers. A. (1997). Flexibleupdate propagation for weakly consistent replication. In Proc. of ACMsymposium on operating systems principles (pp. 288–301).
Saito, Y., & Shapiro, M. (2005). Optimistic replication algorithms. ACM ComputingSurveys (CSUR), 37(1), 42–81.
Santoro, N. (2006). Design and analysis of distributed algorithms. Wiley InterSciencePublisher.
Shapiro, M., & Kemme, B. (2009). Eventual consistency. In Encyclopedia of databasesystems (pp. 1071–1072). Springer-Verlag.
Spiekermann, S. (2008). User control in ubiquitous computing: Design alternatives anduser acceptance. Shaker Verlag.
Taylor N. E.,& Ives. Z. G. (2006). Reconciling while tolerating disagreement incollaborative data sharing. In SIGMOD.
Terry, D. B., Theimer, M. M., Petersen, K., Demers, A. J., Spreitzer, M. J., & Hauser. C. H.(1995). Managing update conflicts in Bayou, a weakly connected replicatedstorage system. In Proc. of the ACM symposium on operating systems principles.
Terry, D. B., Theimer, M. M., Petersen, K., Demers, A. J., Spreitzer, M. J., & Hauser. C. H.(1995). Managing update conflicts in bayou, a weakly connected replicatedstorage system. In Proc. of the ACM symposium on operating systems principles.
Weaver, J. B., IIIa, Thompsonb, N. J., Weavera, S. S., & Hopkins, G. L. (2009).Healthcare non-adherence decisions and internet health information.Computers in Human Behavior, 25(6), 13731380.
Welbourne, J. L., Blanchard, A. L., & Wadsworth, M. B. (2013). Motivations in virtualhealth communities and their relationship to community, connectedness andstress. Computers in Human Behavior, 29(1), 129139.
