S. Poledna. Fault-tolerant real-time systems: The problem of replica determinism. Kluwer Acad. Publishers, 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....with up to four nodes. The addition of the fifth node may lead to sporadic failures. Replica Determinism: If fault tolerance is implemented by the replication of nodes, then the architecture and the nodes must support replica determinism. A set of replicated nodes 8 is replica determinate [49] if all the members of this set have the same externally visible state, and produce the same output messages at points in time that are at most an interval of d time units apart (as seen by an omniscient outside observer) In a fault tolerant system, the time interval d determines the time it ....

....replica determinism and present their results to a voter (physically located at every consumer of the result) that makes a majority decision. A necessary prerequisite for the implementation of active redundancy in the described form is the replica deterministic behavior of the host software [54] [49]. The TTA provides replica determinism at the CNI of a node, but it is up to the host software to ensure replica determinism within the complete node. The three replicas of the host software run synchronously on the three different host computers and produce their output simultaneously (within the ....

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers, 1995.

....be exhausted at once if a single cause affects several or all of those resources. e.g. if the rotor disk of a jet engine disintegrates and the shrapnel severs all of supply lines of the aircraft s redundant hydraulic system [Hughes89] A special type of common cause fault is the generic fault [Yount85, Poledna96]. As the name suggests, it affects system resources that are identical or similar. Such faults are of particular concern in fault tolerant systems that use redundant resources that are identical, e.g. multiple computing nodes that execute the same software on identical hardware. A software flaw ....

Poledna, S.: "Fault-tolerant real-time systems -- the problem of replica determinism", 1996, Kluwer Academic Publ., 147 pp., ISBN 0-7923-9657-X

....as diverging sensor values, differences in execution timing and timeouts etc. Due to these sources diverging outputs may be produced even if the replicated nodes are correct. It is therefore necessary to ensure replica determinism. The problem of replica determinism is treated by S. Poledna in [Pol96]. 2.4.2 Recovery A process that has been off line due to a transient fault has to reintegrate to the system. If the process is replicated the off line process has to update its internal state since it is no longer valid. This can be done using checkpoints with roll back. However this is seldom ....

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers. 1996.

....of commonly used consensus functions (e.g. f midpoint function and f mean function) is provided for such situations. 3.3. 2 Replica determinism The issue of replica determinism is to ensure that there does not exist any unexpected side eoeect that could lead to a divergence among correct replicas [24]. Enforcing replica determinism is a key issue in all fault tolerant distributed systems using task replication. Two main sources of non determinism are possible. Firstly, there can exist replicated sensors that cause a divergence among replicas. As establishing a consensus among the values ....

S. Poledna. Fault-Tolerant Real-Time Systems The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

....Figure 1, will encompass (i) time bounded reliable communication primitives (time bounded point to point communication, time bounded multicast broadcast) drawn respectively on Figure 1 as Rel. Bcast and Rel. Mcast; ii) replication services implementing passive, active and semi active replication [Pol96] iii) consensus service; iv) persistent storage service; v) dependency tracking service [NMT97] and (vi) clock synchronization service (the clock synchronization algorithm designed in [LL88] is drawn on Figure 1) The purpose of the dispatcher is to provide task dispatching mechanisms and ....

S. Poledna. Fault-Tolerant Real-Time Systems -- The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

....from a consistent state [2] after a site stop, since a checkpoint is saved at each communication (see [3] for details) 3.3 Enforcing determinism of replicated computations Let Rep(S) be the set of the transformed patterns of the whole system S by Hydra. System S is said to be deterministic [11] if, given the same initial state for all copies of replicated resources, after each correct execution of g 2 Rep(S) every copy of an output variable Out of g has the same value and the same state changes are made on every copy of a replicated resource. Sources of non determinism. Two sources ....

S. Poledna. Fault-Tolerant Real-Time Systems -- The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

....defined. In this chapter, we present four replication techniques commonly used in fault tolerant systems which are the lock step execution, active replication, semi active replication, and passive replication techniques. 3. 1 Lock Step Execution In the lock step execution replication technique [Pole96] computation is performed on a group of identical hardware processors. Synchronization between all group members is implemented by hardware. All processors in the group are driven by a common clock and computation is performed almost simultaneously at all processors. The group of processors has ....

....The high overhead is due to the tight synchronization and to the complex replica control mechanisms that have to be implemented in order to maintain system consistency. Non deterministic computation requires the use of agreement protocols whenever a decision has to be reached among replicas [Pole96] When using this technique, mechanisms such as dynamic scheduling and task preemption, which are commonly used in most commercial real time operating systems, cannot be used because the computation has to be performed in the same order at all replicas. Another decision problem is also for ....

[Article contains additional citation context not shown here]

S. Poledna. Fault-tolerant real time systems: The problem of replica determinism. Kluwer Academic Publishers, 1996.

....the predictability of the ChorusR3 kernel, used in the Hades prototype, and we have designed a prototype version of the dispatcher to identify its costs. The work under progress concerns transparent generation of fault tolerance software (active, semi active, passive and temporal replication [16]) Given Hades features, i.e. the provision of a range of flexible services deemed at significantly ease and shorten the development of safety critical real time applications, a large real time application from the avionics application domain is planed to be implemented. ....

S. Poledna. Fault-Tolerant Real-Time Systems -- The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

No context found.

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers. 1996.

No context found.

S. Poledna. Fault-tolerant real-time systems: The problem of replica determinism. Kluwer Acad. Publishers, 1996.

No context found.

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers, Boston, Dordrecht, London, 1996.

No context found.

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

No context found.

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers, 1996.

No context found.

S. Poledna. Fault-Tolerant Real-Time Systems: The Problem of Replica Determinism. Kluwer Academic Publishers, 1995.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC