Researches On Strategies Of Real-time Transactions Scheduling With Data Timeliness Maintenance

As the key to the real-time data service, real-time database systems should support transaction timeliness while preserving data freshness. This is challenging due to factors such as the existence of a large quantity of real-time data, unpredictable system workloads and so on. Unfortunately, most of the studies of real-time databases focus on transaction timeliness instead of efficiently maintaining data freshness or satisfying both of them according to users'requirements and system characteristics. The latter two are more critical for real-time data service and are the subjects of this thesis.The temporal validity of real-time data is normally maintained by sensor transactions. A good scheduling algorithm should have high schedulability, generate low update workloads and low run-time cost. Base on the analysis of HH, ML and DS-FP, a new algorithm called deferrable scheduling with time slice exchange (DS-EXC) is proposed. DS-EXC is then extended to multi-processor platforms. DS-EXC with variable execution times is also studied.In several applications, acceptable values of derived objects can be obtained although not all of its dependent objects are updated at the time. Thus the concept of partial recomputation and the scheduling algorithm based on this are proposed. Traditional on-demand algorithms may be seen as a special case of partial recomputation.The problem of scheduling preemptable queries with relative consistency constraints is studied, given that the temporal validity of base data objects has been preserved. The optimal algorithm and some hurestics are presented.In order to maintain data timeliness and transaction timliness under dynamic workloads, a Qos management scheme (UTDS-FC) is proposed which applys data quality adjustment, admission control and feedback control. The multi-class UTDS-FC is then proposed to meet different Qos requirements of different users.The sensor transaction scheduling algorithms available all assume the set of real-time data objects is predefined and fixed, so are not suitable for dynamic environments. In order to attack this problem, a switch model and two mode switch schemes considering the characristics of sensor transaction scheduling alogorithms and real-time data objects are proposed.