Skip to content

Lockstep (computing)

A fault-tolerance method in which redundant computing channels execute the same state transitions in synchrony so their outputs can be compared for error detection or majority-voted for correction.

Version
v1 · 2026-09-28 · History
Domain-specific #
10469
Domain group
Applied Sciences & Engineering
Origin domain
Computer Science & Software Engineering
Subdomain
Fault Tolerant Computing → Computer Science & Software Engineering

Core Idea

Lockstep computing runs redundant processors or channels through the same operations and well-defined state transitions at closely synchronized times. Outputs are compared: two channels expose disagreement, while three or more can support majority selection under an assumed fault model.

The atomic-step model keeps inputs, outputs, and state updates aligned. Time-shifted lockstep may expose transient faults that would strike simultaneous replicas together, but common-mode design, software, clock, or input faults remain a central limit. Redundancy is useful only with justified independence.

Structural Signature

Sig role-phrases:

  • redundant channels. Provide independent executions of the same computation. Constitutive replication. If altered: One processor cannot cross-check itself in lockstep.
  • equivalent input. Feeds comparable data to each channel. Constitutive alignment. If altered: Different inputs make disagreement uninterpretable.
  • synchronized atomic step. Advances channels between matched states. Identity-bearing timing. If altered: Unbounded asynchronous progress is another redundancy scheme.
  • comparator or voter. Detects divergence or selects majority output. Constitutive fault logic. If altered: Duplication without comparison supplies no lockstep detection.
  • fault-independence design. Reduces common-mode failure through separation or time shift. Necessary safety boundary. If altered: Identical replicas can fail identically.

What It Is Not

  • Parallel computing. Are channels doing the same work or partitioned work?
  • Replication. Are state steps tightly synchronized and compared?
  • TMR. Is three-way voting or the broader lockstep method meant?
  • Backup. Does a passive copy execute concurrently?

Scope of Application

Use lockstep for fault-tolerant architectures with channel equivalence, synchronization, comparison, voting, and fault model explicit.

  • Safety-critical control. Detects processor divergence.
  • Avionics and transport. Uses redundant deterministic channels.
  • Industrial systems. Protects stateful controllers.
  • Space systems. Handles radiation-induced faults.
  • Processor design. Implements cores with matched retirement.

Clarity

Parallel means simultaneous work; lockstep means equivalent work at comparable state boundaries. A high-availability replica that lags and later catches up does not necessarily satisfy the timing and atomicity identity.

Manages Complexity

Redundancy converts one execution into a consistency relation among executions. It detects some random faults while multiplying hardware and leaving correlated faults unresolved, so the fault model is part of the architecture.

Abstract Reasoning

  1. Define equivalent channel state and operation sequence.
  2. Align inputs and atomic transition boundaries.
  3. Specify comparison latency and response to disagreement.
  4. Match channel count to detection or correction claims.
  5. Analyze common-mode, software, clock, and input faults separately.

Knowledge Transfer

Replicate–compare reasoning transfers to other dependable systems. Tight state synchronization and voting semantics do not transfer to ordinary backups, consensus services, or distributed replicas with different timing models. The nearest stopping boundary is explicit: Triple modular redundancy is closest and often implemented in lockstep, but TMR names three-way voting while lockstep also includes dual detection and synchronization. The inclusion test remains: A system qualifies when redundant channels execute equivalent operations in synchronized state steps and a comparator or voter acts on disagreement. The structure no longer applies when the case exits when replicas process different work, state alignment is absent, or outputs are never compared.

Examples

Canonical

Three processor channels receive the same sensor frame, complete the same atomic control step, and a voter selects the two matching outputs while isolating one divergent channel.

Mapped back: redundant channels → three processors; equivalent input → same sensor frame; synchronized atomic step → matched control transition; comparator or voter → majority voter; fault-independence design → isolated channels.

Applied / In Practice

Two time-shifted cores execute identical instructions and compare retired state; disagreement detects a transient error, but identical software logic remains a shared failure mode.

Mapped back: redundant channels → two cores; equivalent input → same instruction/data; synchronized atomic step → delayed matched retirement; comparator or voter → divergence detector; fault-independence design → time shift but shared software.

Structural Tensions

T1: tight synchrony vs. fault diversity. Matching timing aids comparison while simultaneous exposure increases common-mode risk. Diagnostic: Which separation improves independence?

T2: correction vs. replica cost. Voting can mask a faulty channel but requires extra resources and a trusted voter. Diagnostic: Which component remains a single point?

Structural–Framed Character

Description turns on redundant channels, equivalent input, synchronized atomic step, comparator or voter, fault-independence design. Skeletal core. Equivalent processes advance together and an agreement relation governs accepted output. Domain-bound accent. Processors, inputs, clocks, atomic state, comparators, voters, and fault models define lockstep. Transfer remains bounded because Why not prime. Redundant agreement is portable; this is a computing architecture. The negative boundary is concrete: Any parallel processing, replication, clustering, backup, deterministic replay, or high availability is not lockstep computing. Lockstep is structural-leaning: replicated state transitions and comparison are formal, while timing and fault independence are engineered. Its character: synchronized redundancy that turns divergence into fault evidence.

Structural Core vs. Domain Accent

Skeletal core. Equivalent processes advance together and an agreement relation governs accepted output.

Domain-bound accent. Processors, inputs, clocks, atomic state, comparators, voters, and fault models define lockstep.

Why not prime. Redundant agreement is portable; this is a computing architecture.

This entry is a kind of Redundancy.

  • Redundancy. Multiple channels provide comparison and masking capacity.
  • Fault detection. Divergence indicates at least one failed execution.
  • No strict parent is asserted.

Relationships to Other Abstractions

Local relationship map for Lockstep (computing)Parents appear above the current abstraction, mutual partners to the right, and children below. Node labels state whether each abstraction is prime or domain-specific; colors identify relation types.Lockstep (computing)DOMAINPrime abstraction: Redundancy — is a kind ofRedundancyPRIME

Current abstraction Lockstep (computing) Domain-specific

Parents (1) — more general patterns this builds on

  • Lockstep (computing) is a kind of Redundancy Prime

    Lockstep computing duplicates computing channels executing identical steps so their outputs can be compared, the redundancy pattern's defining structure.

Hierarchy paths (12) — routes to 8 parentless roots

Neighborhood in Abstraction Space

Lockstep (computing) sits in a sparse region of the domain-specific corpus (69th percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Unclustered & Miscellaneous (2551 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With

  • Parallel computing. Tell: Are channels doing the same work or partitioned work?
  • Replication. Tell: Are state steps tightly synchronized and compared?
  • TMR. Tell: Is three-way voting or the broader lockstep method meant?
  • Backup. Tell: Does a passive copy execute concurrently?

References

  • Frozen Wikipedia discovery revision: https://en.wikipedia.org/wiki/Lockstep_(computing) (revision 1362915942).
  • Preserved source candidate: https://books.google.com/books?id=UVlq7SFDCVUC&q=lockstep+fault+tolerance&pg=PA80
  • Preserved source candidate: https://software.intel.com/en-us/articles/intel-xeon-processor-e7-v2-family-technical-overview#c104
  • Preserved source candidate: https://software.intel.com/en-us/blogs/2014/07/11/independent-channel-vs-lockstep-mode-drive-you-memory-faster-or-safer
  • Preserved source candidate: http://ftp.hp.com/pub/c-products/servers/options/Memory-Config-Recommendations-for-Intel-Xeon-5500-Series-Servers-Rev1.pdf#page=8
  • Preserved source candidate: http://www.intel.com/content/dam/www/public/us/en/documents/datasheets/c102-c104-scalable-memory-buffer-datasheet.pdf#page=9
  • Preserved source candidate: https://docs.nvidia.com/cuda/cuda-programming-guide/01-introduction/programming-model.html#warps-and-simt
  • Preserved source candidate: http://www.dell.com/downloads/global/power/ps3q05-20050176-Patel-OE.pdf
  • Preserved source candidate: https://web.archive.org/web/20150923233016/http://www.ece.umd.edu/courses/enee759h.S2003/references/chipkill.pdf

The frozen Wikipedia revision is discovery provenance. The retained source set was reviewed for identity, formal or operational relation, and scope. The encyclopedia's structural synthesis is bounded to those claims; a thin authority surface is recorded as a nonblocking source-strengthening repair rather than concealed.