Department of Computer Engineering
Kwangjin-Ku, Kunja-Dong 98
Seoul, 143-747, KOREA
Heon Y. Yeom
Department of Computer Science
Seoul National University
Seoul, 151-742, KOREA
Checkpointing, Distributed shared memory, Fault tolerant systems, Message logging, Lazy release consistency.
This paper presents a new causal logging scheme for lazy release consistent distributed shared memory systems. For the efficient implementation of causal logging, data structures and operations supported by the lazy release consistency memory model are utilized. Also, unlike the previous scheme which logs the vector clock for each synchronization operation, the proposed scheme adds the minimum information to recreate the corresponding vector clock, into the existing write notice structures. As a result, the additional information carried in each message becomes less than two integers for each synchronization interval, and hence, fault-tolerance can be achieved with very little overhead. Moreover, the size of the additional information is independent of the number of processes in the system, which means that the new scheme can be very effective for the large size systems. To evaluate the performance of the proposed scheme, the logging protocols have been implemented on top of the CVM(Coherent Virtual Machine), and the experimental results show that the proposed scheme achieves 35% - 66% of reduction in the log size.