Reliable communication in the presence of failures
ACM Transactions on Computer Systems (TOCS)
Update propagation in Bakunin data networks
PODC '87 Proceedings of the sixth annual ACM Symposium on Principles of distributed computing
Preserving and using context information in interprocess communication
ACM Transactions on Computer Systems (TOCS)
An efficient reliable broadcast protocol
ACM SIGOPS Operating Systems Review
Ordered and reliable multicast communication
ACM Transactions on Computer Systems (TOCS)
Lightweight causal and atomic group multicast
ACM Transactions on Computer Systems (TOCS)
Paralex: an environment for parallel programming in distributed systems
ICS '92 Proceedings of the 6th international conference on Supercomputing
Understanding the limitations of causally and totally ordered communication
SOSP '93 Proceedings of the fourteenth ACM symposium on Operating systems principles
IEEE Transactions on Parallel and Distributed Systems
ACM Transactions on Computer Systems (TOCS)
Causality tracking in causal message-logging protocols
Distributed Computing
Hi-index | 0.00 |
In their paper Understanding the Limitations of Causally and Totally Ordered Communication [CS93], David Cheriton (Stanford University) and Dale Skeen (Teknekron Software Systems, Inc.) identify several potential problems in using causally and totally ordered communication support (CATOCS), and conclude that such support is of limited merit at best. This is a remarkable statement, given that, first, many well-known researchers are advocating a CATOCS model, and second, several projects have used CATOCS support successfully.[CS93] lists four limitations. They recognize that the third limitation is only a generalization of the second, so that there are really only three. However, in the course of the paper they state an additional two. For convenience, we combine overhead and scalability concerns. In total, therefore, the paper actually does contain four limitations. These are:• Lamport's event ordering cannot always be recognized in a system, and therefore correct causal delivery cannot always be enforced.• CATOCS delivery is not sufficient to guarantee the consistency of user-level state. The mechanisms that do guarantee such consistency obviate the need for CATOCS.• CATOCS events, even if atomic, are not durable, that is, they may be lost after delivery.• CATOCS presents unacceptable overhead, and does not scale.They conclude that CATOCS is at best very narrowly applicable. These limitations are stated without proof or experimental verification. Here, we undertake to refute their conclusion by showing their examples and proof incorrect. We assume familiarity with the contents of [CS93], and, for the sake of brevity, do not repeat their examples.