Exploiting locality in maintaining potential causality

Abstract

In distributed systems it is often important to be able to determine the temporal relationships between events generated by different processes. An algorithm to determiue such relationships is presented in [3] and [5]. This algorithm has many favorable attributes such as it allows for any kind of interprocess communication, and it requires no extra synchronization messages, additional communication links or cent ral t imestamping authority.The algorithm, however, requires O(n) space for each process (where n is the number of processes).i.e., it requires an overall space of 0(n2 ). This can be a large overhead especially when there are a very large number of processes.By cutting down on this generality, we can significantly decrease the amount of space required to determine temporal relationships.In this paper, we show how one may reduce the space requirements by assuming that the communication links between processes is static and known ahead oj time; and also that one is interested only in determining the temporal ordering between messages arriving at the same process. We argue that these assumptions are reasonable to make for a large class of problems.

Detecting causal relationships in distributed computations: In search of the holy grail

“…However, it suffers from a slight deficiency, as mentioned by Meldal et al in [47]. By compressing the message timestamps, we lose immediate access to some information about the causal relationship between different messages sent to the same receiver.…”

Scalability approaches for causal multicast: a survey

“…Meldal et al (1991) [46] proposed different ways for compacting the vector clocks used in static multi-centre topologies. They proved that when all subgroup interconnection channels are known in advance, vector clocks can be highly compacted.…”

“…In a similar way, Meldal et al [46] proved that the amount of clock-related information to be included in multicast messages can be reduced if the connecting paths are acyclic; also Stephenson (1991) [68] and Mostéfaoui and Raynal (1993) [49] proved that acyclicity is needed. On the other hand, when total-order interconnection is intended, both [68,49] show that an intrusive interconnecting protocol is necessary; i.e., a multicast message cannot be locally delivered in the sending subgroup until the forwarder process has propagated the message and a global total order has been decided.…”

Modeling Time

“…The → relation is referred to as the causality relation by Mattern (1988) since it determines the primary characteristic of time, namely that the future cannot influence the past. It is also referred to as potential causality by Meldal et al (1991) since e → e′ does not necessarily mean that e affects e′, only that e could affect e′.…”