Abstract
The use of distributed systems has become increasingly popular due to increased computing power, reliability, availability, and efficiency in processing. Checkpointing and rollback recovery is a well known technique used to restore a system from failures. Checkpointing schemes have been classified into two categories namely, coordinated checkpointing schemes and independent checkpointing schemes. The objective of this thesis is to compare the performance of both categories. [4] is chosen as the representative for the coordinated schemes and [21 and [3] are chosen for the independent checkpointing. A distributed application is simulated on a subnet of nine SUN Sparc Stations and performance of checkpointing schemes is measured by varying the number of messages, checkpoint size, and checkpoint interval in the application. Analysis of variation (ANOVA) and Tukey-Kramer Comparison Test are performed on the results. We conclude that the coordinated scheme performs equally well as the independent scheme(s) when the number of messages in the system is low. However, with increased number of messages, performance of the coordinated scheme degrades significantly over that of the independent scheme(s). Also, checkpoint size and checkpoint interval do not have a statistically significant influence (p>0.05) on the performance.
Soni, Sameer (1994). Implementaion and performance of checkpointing schemes in a distributed environment. Master's thesis, Texas A&M University. Available electronically from
https : / /hdl .handle .net /1969 .1 /ETD -TAMU -1994 -THESIS -S6986.