An efficient decentralized approach to processor-group membership maintenance in real-time LAN systems: the PRHB/ED scheme

A group membership protocol for communication systems with both static and dynamic scheduling

“…The protocol is also very responsive, removing faulty processors no later than two TDMA cycles after they fail, and (re)integrating them no later than two TDMA cycles after they decide to join. These numbers compare favorably with protocols designed for systems based on TDMA with static scheduling only: the protocol is as fault tolerant as the most fault tolerant protocols [10], and has a slightly higher overhead than the most efficient [7], which assumes only one fault per TDMA cycle.…”

“…The group membership protocol used in TTP/C is essentially the one described by Kopetz and Gruensteidl in [6], and an optimized version of this protocol with respect to failure detection latency is proposed in [7]. The system model used in these works is very similar to ours.…”

“…Essentially, the differences among the published protocols depend on the failure assumptions; usually, the stronger these assumptions, the more efficient the protocol. For example, the group membership protocol [6,7] executed in every cycle of the TTP/C [8], requires the sending of only one additional bit per cycle per processor [9]. However, it can tolerate at most one failure within any interval of two TDMA cycles.…”

Distributed diagnosis in dynamic fault environments

“…A problem closely related to distributed diagnosis, known as the synchronous group membership problem [6], [7], [11], [12], [15], is for each working node to maintain correct information about the group of working nodes with which it can communicate, and for all nodes in one group to agree on the membership of the group. In [12], it is shown that, under some models, these two problems are equivalent and an algorithm for one problem can be converted to an algorithm for the other.…”

Fault detection using hints from the socket layer

“…Process membership services have been developed in the context of group-based systems, and their main objective is to provide a consistent view of which processes are currently members of a group, despite process joints, departures or failures [2][3][4][5][6][7][8]. Distributed system-level diagnosis protocols locate process failures, and then distribute this information in such a way that each node can independently determine the set of failure-free processes [9][10][11][12][13][14].…”