Discover everything there is to know about the scientific topic Recover Command with ideXlab!

Notes on Database System Reliability ∗

We have so far assumed that no system failures occur. Within this context, concurrency control algorithms enforce the isolation property as well as database consistency. The no-failure assumption is, of course, unrealistic; there will be various failures in any computer system. Therefore appropriate protocols have to be implemented such that DBMSreliability can be guaranteed. DBMSreliability protocols address transaction atomicity and durability. As we discussed, atomicity refers to the “all-or-nothing ” property of transactions. Consequently, we need to guarantee that, in the face of failures, either the effects of all of the actions of a transaction are reflected in the database, or none of the effects are. For example, if we are running a transaction that increases the salaries of employees by 10%, and a failure occurs during its execution, the DBMShas to take special care to ensure that the database reflects the increase to the salaries of either all the employees or to none. Durability, on the other hand, requires that the effects of committed transactions on the database survive failures. This also requires special care, since failures may wipe out some of these updates. 1 Architectural Considerations The responsibility for maintaining atomicity and durability of transactions falls on the DBMSRecovery Manager (RM). Applications interface with RM interface using the (abstract) Commands: begin transaction, read, write, commit, and abort 1; RM also has an interface to the operating system via the Recover Command. For ease of exposition, we will assume the system architecture given in Figure 1. This is similar to that given in [HR83] and [BHG87]. In this discussion we assume that the database is stored permanently on secondary storage, which in this context is called the stable storage [LS76]. The stability of this storage medium is due to its robustness to failures. A stable storage device would experience considerably less-frequent failures than would a nonstable storage device. In today’s technology, stable storage is typically implemented by means of duplexed magnetic disks which store duplicate copie