DocumentCode
2593575
Title
Deadlocks in fully uncoordinated checkpointing rollback recovery systems
Author
Shah, Viral ; Sanyal, Sandeepan ; Bhattacharya, Sourav
Author_Institution
Dept. of Comput. Sci. & Eng., Arizona State Univ., Tempe, AZ, USA
fYear
1997
fDate
5-7 Feb 1997
Firstpage
190
Lastpage
197
Abstract
Synchronization issues in checkpointing and rollback recovery schemes have been dealt with in depth over the past few years. The authors investigate the possibility of deadlocks in a fully uncoordinated checkpointing system. A protocol is first illustrated for a fully uncoordinated checkpointing scheme. Rollback propagation analysis (RPA) is performed using a stack based algorithm. The probability of deadlock (due to rollbacks) for a finite buffer size is then computed. The optimal number of buffers required to eliminate the possibility of deadlock is calculated. Finally a comparative analysis is performed between the predicted buffer size and the simulated result. The simulation study shows that the probability of deadlock decreases as the number of buffers increases, till an optimal buffer size is reached where the deadlock probability becomes zero
Keywords
buffer storage; distributed processing; probability; synchronisation; system recovery; virtual machines; deadlock probability; deadlocks; finite buffer size; fully uncoordinated checkpointing rollback recovery systems; optimal buffer size; protocol; rollback propagation analysis; simulation study; stack based algorithm; synchronization issues; Buffer storage; Checkpointing; Computer science; Control systems; Fault detection; Fault tolerant systems; Performance analysis; Protocols; System recovery; Yarn;
fLanguage
English
Publisher
ieee
Conference_Titel
Object-Oriented Real-Time Dependable Systems, 1997. Proceedings., Third International Workshop on
Conference_Location
Newport Beach, CA
Print_ISBN
0-8186-8046-6
Type
conf
DOI
10.1109/WORDS.1997.609953
Filename
609953
Link To Document