Skip to content

Latest commit

 

History

History
179 lines (123 loc) · 5.37 KB

File metadata and controls

179 lines (123 loc) · 5.37 KB

System Threat Resilience & Extreme Stress Suite (STRESS) v0

Reference Workloads

Version: 0.2
Status: Normative Specification
Related Documents:

  • STRESS_v0_Technical_Brief.md
  • STRESS_v0_Complete_Specification.md

1. Purpose and Normative Status

This document defines normative reference workload classes and canonical workloads for STRESS v0.

STRESS results are invalid unless workloads are explicitly defined in accordance with this specification. This document is normative for STRESS v0 workload definition and execution.

Reference workloads provide:

  • A standardized basis for comparative evaluation
  • Consistent interpretation of Behavioral Proxies
  • Reproducibility across implementations and environments

2. Workload Design Principles

All STRESS reference workloads MUST satisfy the following:

  1. Deterministic Baseline Behavior
    Under baseline (unstressed) conditions, workloads MUST complete correctly and reproducibly.

  2. Stress Sensitivity
    Workloads MUST exhibit measurable behavioral change under injected stress.

  3. Observability
    Completion, degradation, and failure MUST be observable via external signals.

  4. Implementation Neutrality
    Workloads MUST NOT assume specific hardware, operating systems, orchestration frameworks, or proprietary runtimes.


3. Workload Declaration Requirements

Each STRESS workload execution MUST include an explicit workload declaration containing:

  • Workload ID and class
  • Input size and distribution
  • Concurrency level
  • Resource limits (CPU, memory, bandwidth)
  • Execution duration or completion criteria
  • Failure classification rules (recoverable vs irreversible)

Failure classification MUST conform to STRESS_v0_Complete_Specification §4.2.


4. Workload Classes (STRESS v0)

STRESS v0 defines three normative workload classes. Results MUST NOT be compared across classes.

4.1 Class W1 — Stateless Task Processing

Description
Independent, stateless tasks with no shared mutable state.

Examples

  • Batch computation over independent inputs
  • Map-style numerical transforms

Baseline Correctness Criteria

  • ≥ 99% task completion
  • Deterministic output verification

Behavioral Proxy Coverage

  • GDS: throughput degradation under stress
  • ARR: recovery from transient execution faults
  • REC: efficiency loss under constraint
  • CFR: expected near-ideal containment

4.2 Class W2 — Stateful Pipeline Processing

Description
Multi-stage workloads maintaining state across execution stages.

Examples

  • Ingest → transform → aggregate pipelines
  • Stream processing with checkpoints

Baseline Correctness Criteria

  • ≥ 99% stage completion
  • Checkpoint integrity preserved
  • Deterministic stage ordering

Behavioral Proxy Coverage

  • GDS: pipeline throughput degradation
  • ARR: recovery from mid-pipeline faults
  • IST: state survival during isolation
  • CFR: failure propagation across stages

4.3 Class W3 — Distributed Coordination Workload

Description
Multiple cooperating components requiring coordination to achieve progress.

Constraints

  • Non-byzantine failure model
  • Fixed participant count (3–7 components)
  • Single coordination primitive per run

Examples

  • Leader election
  • Barrier synchronization
  • Distributed lock acquisition

Baseline Correctness Criteria

  • Safety violations = 0
  • Liveness achieved within declared time bound

Behavioral Proxy Coverage

  • GDS: coordination success rate under stress
  • ARR: recovery from partial participant failure
  • IST: survivability during isolation
  • CFR: containment of participant failure

5. Canonical STRESS v0 Workloads

STRESS v0 defines the following canonical workloads:

ID Class Description
W1-A W1 Stateless batch computation over fixed-size inputs with deterministic verification
W2-A W2 Three-stage pipeline with checkpoint after stage two
W3-A W3 Leader election with heartbeat and timeout parameters

Mandatory Requirement
Implementations claiming STRESS v0 compliance MUST support W1-A.

Cross-implementation comparisons require identical workload IDs and parameters.


6. Measurement Mapping

Workloads MUST expose observable signals sufficient to measure:

  • Task or stage completion rate (GDS)
  • Autonomous recovery events (ARR)
  • Time-to-failure during isolation (IST)
  • Work-per-resource metrics (REC)
  • Failure propagation boundaries (CFR)

Observable signals SHOULD be derivable from external interfaces (e.g., exit codes, logs, counters, timestamps) without privileged internal instrumentation.


7. Extension Rules

  • New workload classes require a major version increment.
  • New workloads within existing classes MAY be added in minor versions.
  • Canonical workload definitions are immutable once versioned.

8. Interpretation Constraints

  • STRESS scores MUST NOT be compared across different workload classes.
  • Workload ID and parameters MUST be disclosed alongside SRI results.
  • Optimizing systems specifically for a workload MUST be disclosed.

9. Summary

Reference workloads ground STRESS’s behavioral metrics in observable execution behavior.

They ensure STRESS evaluates resilience under constraint rather than abstract capability, and that results remain reproducible, interpretable, and comparable across implementations.


End of Reference Workloads