RECOVERY ANALYSIS OF DATA SHARING SYSTEMS UNDER DEFERRED DIRTY PAGE PROPAGATION POLICIES

In a multinode data sharing environment, different buffer coherency co ntrol schemes based on Various lock retention mechanisms can be design ed to exploit the concept of deferring the propagation or writing of d irty pages to disk to improve normal performance. Two types of deferre d write polices are considered. One policy only propagates dirty pages to disk at the times when dirty pages are flushed out of the buffer u nder LRU buffer replacement. The other policy also performs writes at the times when dirty pages are transferred across nodes. The dirty pag e propagation policy can have significant implications on the database recovery time. In this paper, we provide an analytical modeling frame work for the analysis of the recovery times under the two deferred wri te policies. We demonstrate how these policies can be mapped onto a un ified analytic modeling framework. The main challenge in the analysis is to obtain the pending update count distribution which can be used t o determine the average numbers of log records and data I/Os needed to be applied during recovery. The analysis goes beyond previous work on modeling buffer hit probability in a data sharing system where only t he average buffer composition, not the distribution, needs to be estim ated, and recovery analysis in a single node environment where the com plexities on tracking the propagation of dirty pages across nodes and the buffer invalidation effect do not appear. A clipping mechanism can be employed to improve recovery time where the number of pending upda te an a dirty page is limited by forcing a dirty page to disk after th e number of updates accumulated on this page exceeds a certain thresho ld. The analysis captures the effect of clipping also. Finally, we sho w the sensitivities of the recovery time and normal performance to the clipping count.