Deadlock detection in distributed systems

In the literature only a handful of studies have been performed on the distributed deadlock problem in the generalized request model. Most of those algorithms use either the diffusing computation technique or have the initiator collect all the dependency information among processes. This paper proposes an algorithm which incorporates these two methods with the following properties: first, it removes the reduction phase of the diffusing computation; second, it encodes the dependency information to reduce message length, rather than transmitting it naively as is. The main advantage of the proposed algorithm is that deadlock detection time is reduced to almost half of that of the existing algorithms.

Download Full-text

Correct two-phase and one-phase deadlock detection algorithms for distributed systems

Proceedings of the Second IEEE Symposium on Parallel and Distributed Processing 1990 ◽

10.1109/spdp.1990.143519 ◽

2002 ◽

Cited By ~ 8

Author(s):

A.D. Kshemkalyani ◽

M. Singhal

Keyword(s):

Distributed Systems ◽

Deadlock Detection ◽

Two Phase ◽

Detection Algorithms

Download Full-text

An Improved Multi-Cycle Deadlock Detection and Resolution Algorithm for Distributed Systems

Computer Engineering and Applications Journal ◽

10.18495/comengapp.v3i3.89 ◽

2014 ◽

Vol 3 (3) ◽

pp. 111-120

Author(s):

Mohsen Askari ◽

Rozita Jamili Oskouei

Keyword(s):

Distributed Systems ◽

Data Sharing ◽

Deadlock Detection ◽

A Algorithm ◽

System A ◽

Memory Overhead ◽

Resolution Algorithm ◽

High Degree ◽

Available Resources

Distributed systems exhibit a high degree of resource and data sharing creating aÂ state in which deadlocks might make their appearance. Since deadlock detection andÂ resolution Â is one of the important concerns in distributed systems which lead toÂ minimizing available resources, therefore instigating Â the Â system Â throughputÂ decrease. Â Our Â proposed algorithm detects and resolves Â the Â multi-cycle Â deadlocks,Â whether the initiator is involved in the deadlock cycle directly or indirectly. Also theÂ chance Â of Â phantom Â deadlock Â detection is minimized. This Â algorithm Â not Â only Â canÂ manage the simultaneous execution of it but also detects the multi-cycle deadlocksÂ in Â the Â distributed Â systems. Our Â algorithm introduces a modified probe and victimÂ message Â structure. Moreover, Â no Â extra Â storage Â required Â to Â store Â prob message Â inÂ each node which is known as memory overhead in the distributed systems.

Download Full-text