Detection Of Deadlocks In Distributed Database Systems*
Anthony Marsland, Sreekaanth S. Isloor · INFOR Information Systems and Operational Research · 1980
Designers of DBMS have faced many problems with concurrent access and update of information. V/hen a database is distributed over several dissimilar computer systems, such problems as detection and avoidance of process deadlock are compounded. Additional communications may be necessary between the computers to determine the relationships between the data resources and processes.In this paper a thorough discussion of the drawbacks of, and the problems iavolved in, previous proposals to handle deadlocks in distributed DBMS is provided. Earlier algorithms have required substantial communication and coordination between computers should a deadlock arise, thus delaying corrective action. The concept of "on-line" deadlock detection is introduced here; using a graph-theoretic model to represent the process interactions, a new algorithm for the immediate detection of deadlocks in distributed databases is proposed. The theorems upon which our algorithms are based are presented without formal proof. A tutorial approach is taken and application of the theorems is illustrated through a set of examples. Several highlights of our proposals are emphasized.