Fail-stop processor
A fail-stop processor combines halting before faulty state becomes visible, detectable failure and stable storage accessible to other processors.
Schneider's 1984 formulation combines three properties: halt before an erroneous state change becomes visible, allow another processor to detect failure, and preserve readable stable storage. A machine that merely stops responding does not establish all three. [2]
The 1983 treatment separates volatile state from stable state and permits recovery on another processor or on the repaired original processor. Fail-stop therefore should not be silently equated with crash-stop's no-recovery process model. Recovery code must still handle intermediate states left by interrupted work; halting does not make a multi-step update atomic. [1]
Stable storage is a behavioral abstraction, not a synonym for one nonvolatile drive. The papers construct bounded-fault approximations using replicated processors and memory. A k-fail-stop construction guarantees the abstraction only within its stated fault budget; correlated faults and communication assumptions require separate scrutiny. [1] [2]
For a storage design, state the fault containment, detection, surviving-state access and restart contracts separately. A perfect failure detector specifies detection properties, not the historical stable-storage guarantee. [3] [2]
Relations
Limitations
- Different uses of fail-stop may assume different detection and storage contracts. A protocol's explicit fault model determines which guarantees apply. [1, 2]
Citations
- Schlichting and Schneider. Fail-Stop Processors: An Approach to Designing Fault-Tolerant Computing Systems. TOCS 1983, sections 2 and 3.1.
- F. B. Schneider. Byzantine Generals in Action: Implementing Fail-Stop Processors. TOCS 2(2), 145-154, 1984; sections 1-2.
- T. D. Chandra and S. Toueg. Unreliable Failure Detectors for Reliable Distributed Systems. JACM 43(2), 225-267, 1996; sections 1 and 2.1-2.4.
Sources & caveats
- Source synthesis, not a claim that commodity hardware implements an ideal fail-stop processor.
- The 1983 recovery discussion and 1984 three-property formulation are distinguished explicitly; publication dates are not claimed as invention dates.