View Propagation and Inconsistency Detection for Cooperative Mobile Agents

Mobile agents are autonomous programs that migrate from one machine to another within a network on behalf of a client thus they are ideal for mobile computing environments since tasks can be delegated to mobile agents when a mobile client is disconnected This paper extends the traditional functionality of a mobile service agent with capabilities that facilitate asynchronous cooperation among mobile database clients In the context of mobile client server database applications data cached to support disconnected operations can take the form of a materialized view We design mobile agents to reduce computation and wireless com munication costs and use view versioning to cope with disconnected op erations by allowing application sessions to access current data without invalidating work previously done A data validation or results propa gation process detects inconsistencies with newer versions of data upon reconnection Essentially these mobile agents will compute the period of time or consistency window measured in versions for which the re sults of a mobile client s application are consistent We supply rules that govern the creation and sharing of results and show how inconsistencies can be detected to o er a higher availability of data while organizing and gracefully degrading the amount of consistency achieved between the mobile clients and the data sources Customized Query Processing for Mobile Agents In previous work we explored the customization and localization properties of views in the context of mobile database environments to support disconnected query processing and developed a view maintenance mechanism called the View Holder The core of the View Holder is a versioning mechanism that can adjust the currency of the data stored on the mobile client for example by allowing a user who was disconnected during a plane ight to later read updated derived data without necessarily discarding work performed on older data during the ight In addition a view holder is dynamic and stateful with respect to an individual mobile client and therefore it can respond to a mobile client s queries for information by communicating only the di erences between answers thus reducing the cost of wireless communication In contrast to the materialized views maintained by a large static and stateless data warehouse the View Holder can be thought of as a customizable client oriented data warehouse and it does not require modi cations to be made to the existing data sources Delivering the results of queries in a mobile environment is di erent than in a traditional distributed environment due to the rapidly changing conditions of the wireless communication network the requirements of the user in terms of the data s accuracy and the cost the user is willing to pay for communication Traditional query processing facilities are generally concerned with minimizing response time By contrast in a mobile environment a user may want to in troduce delays or change data accuracy in order to save service charges or to minimize required resources Clearly there is a need for devising ways by which mobile users can specify their choices for view maintenance and communication in particular a criteria for materialization that describe which data changes should invoke an update in the view holder agent Instead of using a generic pro le it seems more natural to specify user pref erences along with the de nition of the view to be customized Thus we propose to extend SQL so that the create view statement sent within the create view message includes the view maintenance preferences of a cache agent CA re siding on the mobile device Towards this we introduce the ON condition that can specify which data should be monitored by the view holder agent and how often Essentially the ON condition creates the customizable data currency and summary required by the mobile client s application sessions ASs Customized SQL Statements The extended SQL create view statement o ers additional but optional clauses such as Update On and Maintenance CREATE VIEW name of view AS SELECT attribute list FROM table list WHERE selection and join conditions GROUP BY grouping attribute s HAVING group condition ORDERED BY attribute list UPDATE ON logical expression of pairs condition for materialization Full or Partial MAINTENANCE Recomputational or Incremental The MAINTENANCE clause speci es the view maintenance strategy either recomputational or incremental that should be used by the view holder agent Finally UPDATE ON provides a logical expression of conditions for material ization This generic condition for determining materialization over data servers DSs and data warehouses DWs includes Update ON an individual attribute at DS DS Items price a condition on an attribute DS Items price a maintenance transaction commits at DW DW new transaction a given amount of time has passed minutes or DS minutes This helps the CA when planning a disconnection a logical combination of all the above E g DS Items price OR DW new transaction DS Items price AND DW new transaction Although incremental view maintenance e g can be speci ed the View Holder environment is di erent from the data warehousing environment The amount of customized data requested by mobile clients is orders of magnitude less than what is available from a data warehouse however there can be many more mobile clients and their view holder agents within the xed network Re computational view maintenance methods are more suitable under this scenario since we can perform recomputation without having to capture every update performed at all the data sources because we will not be computing the view during reconstruction Natural periods of disconnection or weak connectivity allow recomputated results to be stored within the view holder agent Recom putational maintenance is easier to implement and incurs fewer latencies since several rounds of queries are no longer required Since small amounts of data are requested the storage space required for intermediate results will also be orders of magnitude smaller than what is required in data warehousing environments Furthermore subqueries will be requested once and only once from each data source during the recomputation Therefore recomputation of the view within the xed network is more appropriate when using DBMS agents since these agents can travel once to each data source transporting results and reconstruct ing a materialized view without having to backtrack and perform compensatory queries at sites already visited Essentially the View Holder approach as shown in Figure allows either recomputational or more complex incremental view maintenance to occur in the xed network while only the view i e incremental maintenance is commu nicated across the expensive wireless links to the cache agent residing on the mobile device A view holder agent computes the variations or view between any two and possibly non consecutive versions of a materialized view The next section will provide an example of how the distributed query processing library routines can construct a materialization program to perform recomputational view maintenance when monitoring loops are used to evaluate the ON condi tion