Abstract
Prioritized atomic multicast consists in delivering messages in total order
while ensuring that the priorities of the messages are considered; i.e., messages with
higher priorities are delivered first. That service can be used in multiple applications.
An example is the usage of prioritization algorithms for reducing the transaction abort
rates in applications that use a replicated database system. To this end, transaction
messages get priorities according to their probability of violating the existing integrity
constraints. This paper evaluates how that abort reduction may be improved varying
the message sending rate and the bounds set on the length of the priority reordering
queue being used by those multicast algorithms.