Advances in computer networks have made possible the development of distributed database management systems (DDBMSs). In DDBMSs which use locking-based concurrency control, the potential for deadlock is increased by the greater concurrency of operations requiring shared data.
To maintain the operational fidelity of any DDBMS with respect to the problem of deadlock, it has been argued that detection and resolution methods may be more advantageous than prevention or avoidance methods. The important aspects and the drawbacks involved in previous deadlock detection protocols in DDBMSs are briefly surveyed.
This thesis is mainly concerned with a distributed deadlock detection mechanism for the detection of deadlocks in DDBMSs, which requires that only LTWF graphs be built and maintained in individual sites, and makes use of a sufficient condition for a global deadkock not to occur in a network. Several advantages and characteristics of this mechanism are described and further research is also discussed.
In addition, this thesis presents the design and implementation of this mechanism, called a Distributed Deadlock Detection System (DDDS) which is composed of two subsystems, LOCALDL and GLOBALDL subsystems.