1. Practical Impact of Group Communication Theory

ions, Specifications and System Models. In the context of abstractions andspecifications, the paper by Hadzilacos and Toueg [1.27] has played an important role:it has contributed in part by motivating subsequent authors to pay more attention to acareful specification of group communication. A new group communication primitive,called generic broadcast [1.38], which takes the message semantics into account, wasalso proposed. The primitive is parametrized by a conflict relation, and ensures thattwo conflicting messages are delivered in the same order by all processes, while non-conflicting messages need not to be ordered. In the context of system models, the failure detector model, initially defined in amodel where processes do not recover after a crash, has been extended to a model withprocess recovery [1.2]. The paper introduces the notion of good process, a process that isalways up or eventually permanently up, and of bad process, a process that is eventuallypermanently down or unstable (i.e., permanently crashing and recovering). The paperalso introduces the failure detectorSu, based on epoch numbers, for solving consensusin the crash-recovery model. The increasing maturity of the field is also witnessed by the comprehensive surveyon group communication specifications by Chockler, Keidar and Vitenberg [1.18]. The“consensus” survey by Barborak, Malek and Dahbura [1.5] may also be mentioned here,even though the paper is written from a system diagnosis perspective, rather than froma distributed computing point of view. Theoretical Results. Two important theoretical results were established in this period.One, by Chandra, Hadzilacos and Toueg, shows that S (or equivalently W) is theweakest failure detector that allows us to solve consensus in an asynchronous system.