SQUIRO: A Framework for Security-Aware Quantum-Classical Scheduling on Kubernetes

Edoardo Giusto; Ignazio Pedone

REVIEW 3 major objections 6 minor 59 references

SQUIRO: A Framework for Security-Aware Quantum-Classical Scheduling on Kubernetes

T0 review · 3 major / 6 minor · reviewed 2026-08-01 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read A hard security mask that excludes any non-compliant node before optimization, combined with global packing, can make regulated hybrid quantum-classical workloads compliant by construction while cutting cost and energy on underloaded cluste

desk verdict The framework's structural security model and coherence-aware backend scoring are the real contributions; the 51%/63% cost/energy savings are a consolidation artifact of a linear active-node cost model, not measured cost or energy. read the letter →

arxiv 2607.16089 v1 pith:Q4LPUYMK submitted 2026-07-17 cs.ET cs.DC

Ignazio Pedone , Edoardo Giusto This is my paper

classification cs.ETcs.DC

keywords schedulingKubernetesquantum-classicalsecurityposturehardconstraintsenergyefficiencybackendselectionCP-SAT

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that hybrid quantum-classical scheduling should treat security requirements as hard, non-negotiable constraints and optimize cost and energy globally across all pending workloads, rather than filtering one pod at a time. Its core claim is that a 'security mask' that excludes any node failing a mandated check makes compliance a property of the scheduler itself, not a downstream audit. On synthetic Kubernetes clusters, the resulting global optimizer is reported to cut infrastructure cost by up to 51% and energy by up to 63% compared with a greedy Kubernetes-style scheduler in underloaded conditions. The paper also claims that quantum backend choice should be circuit-aware, considering coherence margin, queue pressure, and calibration freshness, and shows that this ranking diverges systematically from ranking by raw two-qubit error rate.

What carries the argument

The load-bearing mechanism is the hard security mask, a binary feasibility filter that is computed before any objective is evaluated and that makes non-compliant workload-node pairs impossible to select. Around it, the formulation uses binary active-node variables y_j with linear cost and energy functions C(y)=Σ κ_j y_j and E(y)=T Σ p_j y_j, turning the scheduling problem into a global packing problem solved with a constraint-programming solver. The backend selector adds a second hard-mask stage and a weighted score whose coherence quality factor q_coh = max(0, 1 − circuit_time/min(T1,T2)) penalizes backends whose coherence budget is exceeded by the circuit.

What would settle it

Measure real per-node idle and dynamic power on a live cluster and replay the underloaded experiment: if measured energy savings are far below 63%, the linear active-node energy model is the culprit. Also feed the security mask a node whose attestation certificate is expired but still listed as fresh; if the scheduler accepts it, 'complete compliance' is not actually enforced.

Watch

Extended reading notes

Core claim

The central discovery is that a strict separation between mandatory security constraints and optimizable preferences can be encoded directly into a scheduling instance. SQUIRO's security posture framework computes a mask as_ij = 0 for any workload-node pair that fails a mandatory requirement (PQC support, attestation freshness, TEE presence, compliance floor), and the solver constraint x_ij ≤ as_ij makes any such pair infeasible regardless of objective value. With feasibility settled, the remaining problem is a global packing problem over binary active-node variables y_j, minimizing cost C(y)=Σ κ_j y_j and energy E(y)=T Σ p_j y_j while maximizing admission. On synthetic clusters this yields

Load-bearing premise

The reported cost and energy savings are computed from linear functions that charge every active node a fixed per-hour cost and fixed power draw, so any scheduler that turns off more nodes scores better by construction; the real-world savings depend on whether idle nodes actually dominate energy and cost as assumed.

Editorial extensions

If this is right

Any workload whose security requirements are expressed as hard constraints is guaranteed compliant placement whenever a feasible solution exists; no post-hoc check is needed.
Global consolidation savings exist mainly where there is idle capacity: 51% cost and 63% energy in underloaded clusters, 20–28% in balanced loads, and near zero in packed or overloaded regimes.
The solver's time grows from about 0.3s at 10 nodes to about 21s at 75 nodes, with 60% of 75-node runs exhausting a 30s budget, so the approach as implemented suits small or incrementally-scheduled clusters.
Backend selection that ignores coherence margin and queue pressure can pick a backend that is strictly worse for a given circuit; the two-stage selector changes ranking accordingly.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

If real energy consumption scales with utilization rather than with node activation, the reported 63% energy saving will shrink substantially; the underlying packing benefit may still be real but the headline figure is a property of the linear active-node model.
The 'compliance by construction' claim holds only to the extent that the security metadata feeding the mask (attestation freshness, PQC capability, TEE presence) is accurate and current at scheduling time; stale or spoofed evidence would break the guarantee.
A natural test is to replay the cost-energy experiment on a live Kubernetes cluster with metered energy and cost; the synthetic-instance assumption is the main reason the headline numbers might not reproduce.
The coherence-quality factor is a heuristic proxy (the paper says it does not model transpilation or error mitigation); a stronger test would be noise-model simulation of the top-ranked backend versus the naive-error-rate backend for actual circuit fidelity.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

First: ignore the 51%/63% headline until you check what it is counting. It is counting active-node count in a cost model where cost and energy are linear in the binary y_j variable. The paper's real contribution is structural: separating hard security requirements from optimizable residual risk, and adding coherence margin to backend selection. That part is worth your time.

The USM/SDM formalism is mostly Pinedo with new names—the four-layer taxonomy is fine but not new math. The security mask x_ij ≤ a^s_ij (Eq. 27) is the genuine piece: a single failed check makes placement infeasible regardless of objective. The paper is unusually candid about its own limits—λR(x) is inactive in the main prototype, M(x) is unimplemented, the demand score in the prototype diverges from Eq. (33), and E4 is explicitly not outcome-validated. That honesty earns it credit.

Now the soft spots. E2's cost C(y)=Σκ_j y_j and energy E(y)=TΣp_j y_j are linear in the active-node set. So any optimizer that packs onto fewer nodes wins both metrics by construction. The reported savings are the accounting consequence of 8.4 vs 29.6 active nodes. The paper never states whether inactive nodes are deprovisioned or remain powered on. If they stay powered—typical for on-prem or managed clusters—the economic and energy claims largely disappear. The baseline is also a custom greedy, not stock kube-scheduler bin-packing, so the comparison is weaker than the abstract implies. And the E1 '3.7-point Φij improvement' is R(x) measuring its own objective; not circular, but not independent evidence.

The structural claims hold up. The framework is a reasonable design for hybrid quantum-classical orchestration, and the coherence quality factor (Eq. 26) captures something aggregate error rates miss. But the quantitative evidence is thin: synthetic instances, no released code or data, key parameter values unstated, and the main reported savings rest on a cost model that may not correspond to real infrastructure.

Who should read this: people working on security-aware orchestration or hybrid quantum-classical scheduling. I'd send it to peer review—the structural contribution and the authors' candor give reviewers a clear path—but I'd expect major revision to clarify what the cost/energy metrics actually measure, replace or justify the baseline, and release artifacts. I would not cite the 51%/63% numbers; I would cite the hard-mask/residual-risk separation if I wrote about secure scheduling.

Referee Report

3 major / 6 minor

Summary. The paper proposes SQUIRO, a Kubernetes scheduling framework built on a Unified Scheduling Model (USM) and a six-step Scheduler Design Methodology (SDM). The authors separate hard security feasibility (Eq. 27) from residual-risk optimisation (Eq. 34), add a circuit-aware quantum backend selector with a colocation hierarchy, and report four synthetic experiments: E1 (security-mask compliance), E2 (cost/energy consolidation vs. greedy placement), E3 (CP-SAT scalability), and E4 (backend-rank divergence). The paper is unusually candid: it states that λR(x) is inactive in the main prototype, that M(x) is unimplemented, that the demand score diverges from Eq. (33), and that E4 has no outcome validation.

Significance. If the claims are accepted, the main contribution is a clean architectural separation between non-negotiable security requirements and gradational security preferences, expressed as a hard constraint rather than a score term; the global-packing result also gives a useful bound on consolidation headroom in underloaded clusters. The hard-mask construction (x_ij ≤ a^s_ij) genuinely makes any feasible schedule compliant with the modelled checks by construction. The paper is also commendable for explicitly stating its own implementation gaps. However, the headline cost/energy savings are computed under a binary active-node accounting model and therefore support a consolidation-accounting claim, not a measured operational-energy or infrastructure-cost claim. The residual-risk E1 result is, as reported, a consistency check on the solver rather than evidence about security value. The overall framework is a plausible design blueprint, but the quantitative headline claims need to be reframed or supported with additional measurement/modeling before publication.

major comments (3)

[§VI B, Eqs. (19)–(20); §VII E2] The 51% cost and 63% energy savings are not measured infrastructure cost/energy savings; they are algebraic consequences of the model C(y)=Σκ_j y_j and E(y)=TΣp_j y_j. Since objective (21) minimizes αC+βE, any schedule activating fewer nodes scores better on both metrics by construction. In E2 underloaded, CP-SAT-balanced uses 8.4 active nodes vs. 29.6, and the reported savings are the direct accounting effect of that difference. The paper never states whether inactive nodes are deprovisioned/powered off, nor models idle power, load-dependent power, or power-state transition costs. If inactive nodes remain powered on, the savings largely vanish. Please either rephrase the abstract and conclusions as consolidation gains under a linear active-node model, or add sensitivity analysis with realistic power/cost curves and explicit node-lifecycle assumptions.
[§VI B residual-risk note; §VII E1; Eq. (34)] The E1 'hard+soft' comparison does not validate the security fit score. R(x) is defined as Σ(1−Φ_ij/100)x_ij, and the reported 'mean Φ_ij rises from 81.0 to 84.7' is simply the solver increasing its own objective. That is a tautology, not evidence that the residual-risk term improves placement quality. The experiment only shows that CP-SAT can optimize the function it is given. Please replace or supplement the comparison with a baseline such as random feasible placement, or with an independent held-out security metric, and soften the conclusion's claim of a 'measurable 3.7-point security fit-score gain'.
[§VI B; §VII E2; Abstract] The abstract's claim 'without compromising admission priorities' is not supported. Section VI B states that 'the current CP-SAT prototype uses a uniform admission reward, with π_i held constant across workloads,' and E2's admission row is reported for CP-SAT-performance, not for a scheduler with differentiated priorities. No experiment uses workload-specific priority weights, so the paper cannot claim that admission priorities are preserved. Please either implement and test differentiated priorities, or reframe the claim as 'without reducing aggregate admission rate'.

minor comments (6)

[§VI C] Typo in connectivity class description: 'hea vy-hex' should be 'heavy-hex'. Also, the sentence describing the omitted closed-loop pair from Fig. 5(d) should be in the figure caption or experiment setup rather than only in the results paragraph.
[§VI D, Eq. (33)] The demand-score implementation is described as an additive pass with an unspecified 'compression factor' that diverges from the closed-form Eq. (33). The paper acknowledges this, but for reproducibility the exact formula should be provided (or Eq. (33) marked as aspirational).
[Fig. 5(b)] The caption mixes schedulers: the plot is labeled 'CP-SAT-balanced over K8s-greedy' but the admission row refers to 'CP-SAT-performance'. Clarify which scheduler is used for each row to avoid confusion.
[§VI B] The unimplemented M(x) locality/disruption term is listed in the objective (21) but never used in any experiment. This is disclosed, but the paper would be clearer if the evaluation explicitly stated that all experiments set δ=0.
[§VII E4 and Conclusion] E4 demonstrates divergence between SQUIRO's score and naive error-rate ranking, but it does not show that the selected backend yields better fidelity or time-to-solution. The conclusion's phrase 'ranking backends by raw error rate alone is insufficient' is stronger than the evidence; it should be softened to 'can diverge under coherence- and queue-limited conditions' unless outcome validation is added.
[Data Availability] The result files are available 'upon reasonable request' but the experimental tooling is not released. Given that all four quantitative experiments depend on synthetic instances, releasing the generator and solver configuration would substantially increase reproducibility.

Circularity Check

1 steps flagged · score 4.0 of 10

E1's residual-risk gain is a self-definitional consequence of the objective; the hard-mask compliance result and E2/E4 are not circular in the same way.

self definitional [§VI B Eq. (21), §VI D Eq. (34), §VII E1 and Fig. 5(a)]
"Residual risk is derived from the fit score: rij = 1 − Φij/100, R(x) = X i∈W X j∈R rij xij. ... The policy-governed scalar objective is min αC(y)+βE(y)+λR(x)+δM(x)−γA(z). ... Activating the soft term raises mean Φij from 81.0 to 84.7 (Wilcoxon W=0, p=1.9×10−9, r=1.00) at a 3.4% cost and 4.7% energy premium: the residual-risk gradient among feasible placements is informative."

The soft term λR(x) is defined via Eq. (34) as the complement of the fit score Φij. Minimizing λR(x) in Eq. (21) is therefore exactly maximizing Σ Φij xij. E1 compares the λ>0 arm to the λ=0 arm and reports that mean Φij rises by 3.7 points. That rise is the optimizer improving the same objective term that defines the reported outcome; it is forced by construction rather than an independent measurement. Only the accompanying 3.4% cost premium is an independent trade-off result.

full rationale

The hard security enforcement (Eq. 27, x_ij ≤ a^s_ij) is a genuine hard constraint; claiming compliance by construction there is a formal property, not circular. The E2 cost/energy savings are computed with the paper's linear active-node model (Eqs. 19-20) and thus are model-validity limited, but they compare CP-SAT against greedy under the same metric and are not definitionally identical to the inputs. E4 explicitly says divergence 'demonstrates that the score responds to coherence and queue conditions as designed' and does not independently validate fidelity, so it is not overclaimed. Self-citations ([34], [35], [45]) are background and not load-bearing. The only clear circular element is E1's fit-score gain, which reduces to the objective λR(x) by Eq. (34) and Eq. (21).

Assumptions & free parameters 8 free parameters · 6 assumptions · 4 invented entities

The quantitative claims rest on: policy-chosen weights whose values are not reported (objective weights in Eq. 21; backend-score weights in Eq. 24; coherence coefficients in Eq. 25; security-dimension weights ωh; fit-score weights νM,νS and penalty rate θ; calibration-freshness defaults), a prototype demand score that diverges from the stated formula (Eq. 33), an energy model linear in active nodes (Eq. 20), calibrated synthetic experiments (E1), and unvalidated heuristic security/fit scores. The structural constraint claims reduce to standard scheduling math; the security-score claims are domain assumptions with no external validation.

free parameters (8)

Objective weights α, β, λ, δ, γ (Eq. 21) = not reported
Policy parameters chosen per 'profile'; the E2 savings numbers measure the result of the chosen α/β/γ balance, but the values used in the reported runs are never stated.
Backend score weights w_Q, w_L, w_V, w_C, w_F and priority multiplier ρ_q (Eq. 24) = not reported
These determine the E4 rankings; without the values the divergence from naive error-rate ranking cannot be reproduced or audited.
Coherence quality coefficients c1..c4 (Eq. 25) = not reported (only 'sum to one')
Balance the median-error terms against the coherence factor; the E4 coherence-driven demotions depend on their relative size.
Calibration-freshness thresholds (24h batch, 6h closed-loop, 2h near-time) = 24h / 6h / 2h
Implementation defaults. In E4 they determine which backends survive the hard filter; combined with the all-remote colocation mapping they make near-time infeasible by construction.
Security dimension weights ω_h (Eqs. 29–30, 33) = not reported
Workload-configurable weights that define Sj, Mij, and Di; E1's 10-check curated subset uses unspecified weights.
Fit-score weights ν_M, ν_S and penalty rate θ (Eqs. 31–32) = not reported
Govern Φij and hence the residual-risk objective λR(x) that E1 claims improves mean Φij by 3.7 points.
Demand-score 'compression factor' = undisclosed
Prototype computes Di with an additive pass plus a compression factor that the text does not specify; the implemented score differs from stated Eq. (33).
E1 node-pool sizing margins = demand share + margin
The experiment is 'deliberately calibrated so a genuine, non-degenerate feasible region exists'; the margin is a hand-chosen quantity that guarantees the discriminating CSR result.

assumptions (6)

domain assumption Security posture facts (PQC algorithm presence, TEE type and level, attestation freshness, TLS/PQC floor, AI-governance declaration) are accurately observable node attributes at scheduling time
§VI D: the hard mask a^s_ij is computed from 'node-level configuration facts extractable from attestation evidence or Kubernetes API state'; if attestation metadata is absent or not trusted, the mask silently misfires.
domain assumption The Φij fit score (Eqs. 30–34) faithfully ranks residual security exposure
Residual-risk optimization's value claim (E1's 3.7-point Φij gain) presupposes the score correlates with real security outcomes; the paper presents no such validation ('six operator-facing heuristic labels, not NIST assurance levels').
domain assumption The coherence-budget ratio (d1q·t1q + d2q·t2q)/min(T1,T2) approximates circuit coherence failure (Eq. 26)
Stated by the authors as 'a heuristic ranking proxy, not a validity test' that ignores transpilation, parallel gates, dynamical decoupling, and error mitigation (§VI C).
domain assumption Synthetic cluster/backend instances generated 'from distributions grounded in published specifications' are representative of real deployments
All quantitative results (E1–E4) come from synthetic data; the paper acknowledges this replaces live-cluster validation (§VII).
domain assumption CP-SAT solutions within the 10–30s budgets are of adequate quality for the reported claims
E3 shows 40–60% of runs exhaust the budget at 25–75 nodes; E2 uses 10s budgets at up to 40 nodes, so the 'savings' may come from best-found rather than optimal solutions.
standard math Capacity, precedence, and capability constraint semantics (Eqs. 8–10)
Standard scheduling feasibility conditions (resource limits, dag precedence with transfer delay, capability coverage) treated as background math.

invented entities (4)

Colocation class hierarchy (remote ≺ network-near ≺ node-near ≺ controller-tight)
purpose: Execution-mode feasibility filter and forward-compatible QPU integration contract (§VI C)
A pure design abstraction; because all current cloud backends map to 'remote', near-time workloads are excluded in E4 by construction, validating the filter rather than any physical constraint (the paper says exactly this).
Coherence quality factor qcoh
purpose: Circuit-specific quality term in backend scoring (Eq. 26)
An acknowledged heuristic ('not a validity test'); its failure mode (coherence-limited deep circuits) motivates E4's demotions, but no fidelity or time-to-solution outcome is measured.
Security dimension / posture / fit score system (Sjh, Sj, Mij, Di, Φij)
purpose: Quantify node security posture, workload demand, and workload-fit residual risk (§VI D)
The scores are operator-facing heuristics; there is no evidence they track real compliance or breach outcomes, so the residual-risk gradient is internally defined.
USM four-layer decomposition (problem/model, architecture/control, algorithmic, implementation)
purpose: Organizing formalism for the SDM design methodology (Fig. 1)
A taxonomy, not a testable entity; it adds vocabulary but no falsifiable handle.

how reviews work

0 comments

Cite this review

Pith. "Pith review of SQUIRO: A Framework for Security-Aware Quantum-Classical Scheduling on Kubernetes." pith.science (2026). https://pith.science/paper/Q4LPUYMK

@misc{pith2026260716089,
  author       = {Pith},
  title        = {Pith review of: SQUIRO: A Framework for Security-Aware Quantum-Classical Scheduling on Kubernetes},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Q4LPUYMK}},
  note         = {Machine review of arXiv:2607.16089}
}

read the original abstract

Distributed infrastructure schedulers traditionally optimise capacity, locality, and cost, but provide limited support for security posture and emerging quantum-classical workloads. As hybrid quantum-classical computing becomes increasingly practical and post-quantum security requirements begin to affect infrastructure deployment, schedulers must jointly reason about heterogeneous compute resources, security constraints, and quantum backend characteristics. We present SQUIRO, a framework for security-aware quantum-classical scheduling based on a platform-independent Unified Scheduling Model (USM) and a six-step Scheduler Design Methodology (SDM) that together enable the derivation of concrete schedulers for Kubernetes, high-performance computing (HPC), and federated environments. The framework combines multidimensional security posture enforcement through hard feasibility constraints with residual-risk optimisation, and introduces a circuit-aware quantum backend selector that accounts for coherence margin, calibration freshness, queue pressure, and hardware capabilities through a forward-compatible colocation hierarchy. Evaluation on synthetic Kubernetes clusters shows that the security model enforces complete compliance for regulated workloads by construction, while global optimisation reduces infrastructure cost by up to 51% and energy consumption by up to 63% compared with greedy placement in underloaded scenarios, without compromising admission priorities. Additional experiments characterise the solve-time growth of the current CP-SAT formulation and show that circuit-aware backend selection systematically diverges from naive error-rate ranking under coherence- and queue-limited conditions.

Figures

Figures reproduced from arXiv: 2607.16089 by the authors.

**Figure 2.** FIG. 2: Roadmap of the staged evolution of SQUIRO, from the already implemented Kubernetes (K8s) and [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** FIG. 3: SQUIRO centralised optimisation architecture. [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 5.** Figure 5: FIG. 5: Preliminary evaluation results (synthetic clusters). [PITH_FULL_IMAGE:figures/full_fig_p015_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

59 extracted references · 1 canonical work pages

[1]

National Institute of Standards and Technology, Migra- tion to post-quantum cryptography,https://pages.nist .gov/nccoe-migration-post-quantum-cryptography/ (), accessed: 2026-05-23

2026
[2]

Instantiate the USM.The analytical output of step 1 is mapped onto the model components, con- structing GW, GR, C, P, O, and a preliminary St and A for the target scope, both finalised in step 3 once the control architecture fixes their ob- servation and control boundary. Domain-specific re- quirements, including security constraints, backend eligibility ...
[3]

Select the control architecture.The choice of control architecture, whether centralised, hier- archical, federated, or decentralised, is determined by the state visibility and authority established in step 1. This decision fixes the control boundary and determines how state is observed and how resource management decisions propagate through the sys- tem, ...
[4]

Derive the mathematical problem and select an algorithm.Based on the outputs of steps 1 through 3, the induced scheduling problem is clas- sified and one or more algorithmic approaches are identified. The problem class may correspond to general allocation, multidimensional bin-packing, job-shop scheduling, or minimum-cost flow, depend- ing on the structur...
[5]

Define the implementation and data model. The abstract model components are translated into concrete resource and workload descriptors, schemas, and platform APIs, mapping abstract re- quirements unambiguously onto the computing re- sources and domain constraints of the target envi- ronment. Extensions such as finer-grained configura- tion, compliance met...
[6]

The first is to surface cross- layer interactions that are not apparent when layers are treated as independent components

Validate, iterate, and evolve.Validation pur- sues two distinct aims. The first is to surface cross- layer interactions that are not apparent when layers are treated as independent components. The second is to resolve conflicts between the abstract model and practical implementation constraints: platform or domain limitations discovered during implementa-...
[7]

R. Fang, R. Bindu, A. Gupta, and D. Kang, LLM agents can autonomously exploit one-day vulnerabilities (2024), arXiv:2404.08144 [cs.CR]

arXiv 2024
[8]

Anthropic,Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign, Tech. Rep. (Anthropic, 2025)

2025

Show all 59 references

[9]

Giusto, S

E. Giusto, S. Núñez-Corrales, A. Cilardo, N. Mazzocca, and T. Humble, A typology of quantum-classical faults, in 2025 55th Annual IEEE/IFIP International Conference on Dependable Systems and Networks Workshops (DSN- W)(2025) pp. 188–195

2025
[10]

National Institute of Standards and Technology, NIST releases first 3 finalized post-quantum encryption stan- dards, https://www.nist.gov/news-events/news/202 4/08/nist-releases-first-3-finalized-post-quant um-encryption-standards(), accessed: 2026-05-23

2026
[11]

Office of Management and Budget, M-26-15: Execution of the migration to post-quantum cryptography,https: //www.whitehouse.gov/wp-content/uploads/2026/06 /M-26-15-Execution-of-the-Migration-to-Post-Qua ntum-Cryptography.pdf, accessed: 2026-07-14

2026
[12]

Kubernetes, Kubernetes scheduling framework, https: //kubernetes.io/docs/concepts/scheduling-evictio n/scheduling-framework/(), accessed: 2026-05-21

2026
[13]

Narayanan, K

D. Narayanan, K. Santhanam, F. Kazhamiaka, A. Phan- ishayee, and M. Zaharia, Heterogeneity-Aware cluster scheduling policies for deep learning workloads, in14th USENIX Symposium on Operating Systems Design and Implementation (OSDI 20)(USENIX Association, 2020) pp. 481–498

2020
[14]

Isard, V

M. Isard, V. Prabhakaran, J. Currey, U. Wieder, K. Tal- war, and A. Goldberg, Quincy: fair scheduling for dis- tributed computing clusters, inProceedings of the ACM SIGOPS 22nd Symposium on Operating Systems Princi- ples, SOSP ’09 (Association for Computing Machinery, New York...

2009
[15]

I. Gog, M. Schwarzkopf, A. Gleave, R. N. M. Watson, and S. Hand, Firmament: Fast, centralized cluster scheduling at scale, in12th USENIX Symposium on Operating Sys- tems Design and Implementation (OSDI 16)(USENIX Association, Savannah, GA, 2016) pp. 99–115

2016
[16]

Pehlivanoglu, U

S. Pehlivanoglu, U. de Muelenaere, P. Kogge, and A. Sabry, Qurator: Scheduling hybrid quantum-classical workflows across heterogeneous cloud providers (2026), arXiv:2604.05505 [cs.DC]

2026 arXiv
[17]

Seitz, M

P. Seitz, M. Geiger, C. Ufrecht, A. Plinge, C. Mutschler, D. D. Scherer, and C. B. Mendl, SCIM MILQ: An HPC quantum scheduler, in2024 IEEE International Confer- ence on Quantum Computing and Engineering (QCE), Vol. 02 (2024) pp. 292–298

2024
[18]

Giortamis, F

E. Giortamis, F. Romao, N. Tornow, D. Lugovoy, and P. Bhatotia, Qonductor: A cloud orchestrator for quan- tum computing, inProceedings of the International Con- ference for High Performance Computing, Networking, Storage and Analysis, SC ’25 (Association for Computing Machiner...

2025
[19]

Al-Haj, E

S. Al-Haj, E. Al-Shaer, and H. V. Ramasamy, Security- aware resource allocation in clouds, in2013 IEEE Inter- national Conference on Services Computing(2013) pp. 400–407

2013
[20]

Felemban, A

M. Felemban, A. Almutairi, and A. Ghafoor, Risk-aware sensitive property-driven resource management in cloud datacenters (2025), arXiv:2502.02720 [cs.CR]

2025 arXiv
[21]

M. L. Pinedo,Scheduling: Theory, Algorithms, and Sys- tems, 5th ed. (Springer, 2016)

2016
[22]

Erata, C

F. Erata, C. Xu, R. Piskac, and J. Szefer, Quantum circuit reconstruction from power side-channel attacks on quantum computer controllers, IACR Transactions on Cryptographic Hardware and Embedded Systems2024, 735–768 (2024)

2024
[23]

Tejedor, B

M. Tejedor, B. Casas, J. Conejero, A. Cervera-Lierta, and R. M. Badia, Orchestrating quantum-hpc workflows with distributed quantum circuit cutting, inProceedings of the SC ’25 Workshops of the International Conference for High Performance Computing, Networking, Storage and An...

2025
[24]

Ramsauer and W

R. Ramsauer and W. Mauerer, Towards system-level quantum-accelerator integration, in2025 IEEE Inter- national Conference on Quantum Computing and Engi- neering (QCE), Vol. 02 (2025) pp. 15–20

2025
[25]

Kubernetes, Device plugins,https://kubernetes.io/do cs/concepts/extend-kubernetes/compute-storage-n et/device-plugins/(), accessed: 2026-05-23

2026
[26]

A. J. McCaskey, D. I. Lyakh, E. F. Dumitrescu, S. S. Powers, and T. S. Humble, XACC: a system-level soft- ware infrastructure for heterogeneous quantum–classical computing*, Quantum Science and Technology5, 024002 (2020)

2020
[27]

Shehata, P

A. Shehata, P. Groszkowski, T. Naughton, M. Gopalakr- ishnan Meena, E. Wong, D. Claudino, R. Ferreira da Silva, and T. Beck, Bridging paradigms: Designing for hpc-quantum convergence, Future Generation Computer Systems174, 107980 (2026)

2026
[28]

T. Beck, A. Baroni, R. Bennink, G. Buchs, E. A. C. Pérez, M. Eisenbach, R. F. da Silva, M. G. Meena, K. Got- tiparthi, P. Groszkowski, T. S. Humble, R. Landfield, K. Maheshwari, S. Oral, M. A. Sandoval, A. Shehata, I.-S. Suh, and C. Zimmer, Integrating quantum computing re- so...

2024
[29]

Saurabh, S

N. Saurabh, S. Jha, and A. Luckow, A conceptual ar- chitecture for a quantum-hpc middleware, in2023 IEEE 19 International Conference on Quantum Software (QSW) (2023) pp. 116–127

2023
[30]

Mantha, F

P. Mantha, F. J. Kiwit, N. Saurabh, S. Jha, and A. Luckow, Pilot-quantum: A middleware for quantum- hpc resource, workload and task management, in2025 IEEE 25th International Symposium on Cluster, Cloud and Internet Computing (CCGrid)(2025) pp. 01–10

2025
[31]

Mantha, F

P. Mantha, F. J. Kiwit, N. Saurabh, S. Jha, and A. Luckow, Hybrid quantum-hpc middleware systems for adaptive resource, workload and task management (2026), arXiv:2604.03445 [quant-ph]

2026 arXiv
[32]

Bacher, M

U. Bacher, M. Birmingham, C. D. Carothers, A. Damin, C. D. G. Calaza, A. K. Karnad, S. Mensa, M. Moreau, A. Nober, M. Ohtani, M. Rossmannek, P. Rubin, M. E. Sahin, O. Wallis, A. Shehata, I. Sitdikov, and A. Wen- nersteen, Quantum resources in resource management systems (2025)...

2025
[33]

Burgholzer, M

L. Burgholzer, M. Walter, P. Hopf, Álvaro Caride- Tabarés Sánchez, T. Mattsson, B. Hoffmann, N. Färkkilä, D. Bulmash, R. Wille, and E. Mansfield, Practical HPCQC integration with QDMI: A real-hardware case study with IQM systems (2026), arXiv:2604.19869 [quant- ph]

2026 arXiv
[34]

Wille, L

R. Wille, L. Schmid, Y. Stade, J. Echavarria, M. Schulz, L. Schulz, and L. Burgholzer, QDMI - quantum device management interface: Hardware-software interface for the munich quantum software stack, in2024 IEEE Inter- national Conference on Quantum Computing and Engi- neering (...

2024
[35]

Shehata, B

A. Shehata, B. Austin, T. Beck, L. Burgholzer, A. Chernoguzov, S. Churchill, A. Delgado, Y. Eckert, J. Heckey, K. Kissell, K. Klymko, J. Moles, T. Naughton, L. J. O’Riordan, C. O. Pauyac, G. Prawiroatmodjo, E. Rrapaj, J. Schindler, L. Schulz, S. Stern, T. Takeshita, M. Tsuji, ...

2026 arXiv
[36]

Stirbu, O

V. Stirbu, O. Kinanen, M. Haghparast, and T. Mikkonen, Qubernetes: Towards a unified cloud-native execution platform for hybrid classic-quantum computing, Informa- tion and Software Technology175, 107529 (2024)

2024
[37]

Tejedor, M

M. Tejedor, M. Grossi, C. Tüysüz, R. Rocha, and S. Vallecorsa, Kubernetes-orchestrated hybrid quantum- classical workflows (2026), arXiv:2603.24206 [quant-ph]

2026
[38]

Weder, J

B. Weder, J. Barzen, M. Beisel, F. Bühler, D. Georg, F. Leymann, and L. Stiliadou, Qunicorn: A middleware for the unified execution across heterogeneous quantum cloud offerings, in2025 IEEE/ACM International Work- shop on Quantum Software Engineering (Q-SE)(2025) pp. 17–24

2025
[39]

Pedone and A

I. Pedone and A. Lioy, Quantum key distribution in ku- bernetes clusters, Future Internet14, 10.3390/fi14060160 (2022)

2022 doi
[40]

Pedone, A

I. Pedone, A. Atzeni, D. Canavese, and A. Lioy, Toward a complete software stack to integrate quantum key dis- tribution in a cloud environment, IEEE Access9, 115270 (2021)

2021
[41]

Seelam, J

S. Seelam, J. M. Chow, A. Córcoles, S. Sheldon, T. Mittal, A. Kandala, S. Dague, I. Hincks, H. Horii, B. Johnson, M. Le, H. Jamjoom, and J. M. Gambetta, Reference architecture of a quantum-centric supercomputer (2026), arXiv:2603.10970 [cs.ET]

2026
[42]

Alexeev, M

Y. Alexeev, M. Amsler, M. A. Barroca, S. Bassini, T. Bat- telle, D. Camps, D. Casanova, Y. J. Choi, F. T. Chong, C. Chung,et al., Quantum-centric supercomputing for materials science: A perspective on challenges and future directions, Future Generation Computer Systems160, 666 (2024)

2024
[43]

S. S. Tannu and M. K. Qureshi, Not all qubits are created equal: A case for variability-aware policies for nisq-era quantum computers, inProceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems,ASPLOS ’1...

2019
[44]

Cipollini, S

M. Cipollini, S. Rizzo, S. Iserte, P. Viviani, G. Vitali, M. Barbieri, G. Bettonte, E. Boella, F. Ganz, R. Rocco, O. Spina, A. J. Peña, P. Sandås, I. Colonnelli, A. Scionti, C. Vercellino, E. Dri, J. Frassineti, S. Marzella, A. Mura- tori, D. Ottaviani, O. Terzo, B. Montrucchi...

2026
[45]

Choudhury, A

N. Choudhury, A. S. Bhave, and K. Basu, Quartet : Quan- tum utilization and adaptation via resource-tuned execu- tion techniques, in2025 IEEE International Conference on Quantum Computing and Engineering (QCE), Vol. 01 (2025) pp. 709–719

2025
[46]

Z. Yang, C. Tian, R. Jain, R. Kompella, R. Nejabati, M. Hamdi, A. Erbad, and H. Shapourian, Effective scheduling for quantum data centers, in2025 IEEE In- ternational Conference on Quantum Computing and En- gineering (QCE), Vol. 02 (2025) pp. 454–455

2025
[47]

Giusto, S

E. Giusto, S. Núñez-Corrales, K. N. Smith, P. Cao, E. You- nis, P. Rech, F. Vella, B. Baheri, A. Cilardo, B. Mon- trucchio, W. Jiang, S. Xu, S. Dasgupta, R. K. Iyer, and T. S. Humble, Dependable classical-quantum computing systems engineering, Frontiers in Computer Science7, 1...

2025
[48]

Trochatos, C

T. Trochatos, C. Xu, S. Deshpande, Y. Lu, Y. Ding, and J. Szefer, Trusted execution environments for quantum computers, Frontiers in Computer Science7, 1521059 (2025)

2025
[49]

A. A. Arellano, H. Xie, and J. Szefer, Qubitvise: Double- sided crosstalk attack in superconducting quantum com- puters, in2025 IEEE International Conference on Quan- tum Computing and Engineering (QCE), Vol. 02 (2025) pp. 50–53

2025
[50]

Chiavassa, A

P. Chiavassa, A. Marchesin, I. Pedone, M. F. Dacrema, and P. Cremonesi, Virtual network function embedding with quantum annealing, in2022 IEEE International Con- ference on Quantum Computing and Engineering (QCE) (2022) pp. 282–291

2022
[51]

Kubernetes, Scheduler configuration, https://kubern etes.io/docs/reference/scheduling/config/ (), ac- cessed: 2026-05-23

2026
[52]

Google OR-Tools, CP-SAT solver,https://develope rs.google.com/optimization/cp/cp_solver , accessed: 2026-05-23

2026
[53]

National Institute of Standards and Technology, The NIST cybersecurity framework 2.0,https://www.nist.g ov/cyberframework(), accessed: 2026-05-23

2026
[54]

Accessed: 2026-05-23

National Institute of Standards and Technology, Security and privacy controls for information systems and organi- zations, https://csrc.nist.gov/publications/detai l/sp/800-53/rev-5/final (), NIST Special Publication 20 800-53 Revision 5. Accessed: 2026-05-23

2026
[55]

National Institute of Standards and Technology, Artificial intelligence risk management framework (AI RMF 1.0), https://www.nist.gov/itl/ai-risk-management-fra mework(), accessed: 2026-05-23

2026
[56]

OWASP Foundation, OWASP top 10 for large language model applications, https://owasp.org/www-project -top-10-for-large-language-model-applications/ , accessed: 2026-05-23

2026
[57]

Google Cloud, Compute Engine Pricing: us-central1, https://cloud.google.com/compute/all- pricing , accessed: 2026-05-23

2026
[58]

IBM Quantum, IBM Quantum Platform, https://qu antum.ibm.com/ , calibration data accessed via Qiskit Runtime, 2026-05-23

2026
[59]

IonQ, IonQ Quantum Computing Systems,https://io nq.com/technology, accessed: 2026-05-23

2026

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.

[1] [1]

National Institute of Standards and Technology, Migra- tion to post-quantum cryptography,https://pages.nist .gov/nccoe-migration-post-quantum-cryptography/ (), accessed: 2026-05-23

2026

[2] [2]

Instantiate the USM.The analytical output of step 1 is mapped onto the model components, con- structing GW, GR, C, P, O, and a preliminary St and A for the target scope, both finalised in step 3 once the control architecture fixes their ob- servation and control boundary. Domain-specific re- quirements, including security constraints, backend eligibility ...

[3] [3]

Select the control architecture.The choice of control architecture, whether centralised, hier- archical, federated, or decentralised, is determined by the state visibility and authority established in step 1. This decision fixes the control boundary and determines how state is observed and how resource management decisions propagate through the sys- tem, ...

[4] [4]

Derive the mathematical problem and select an algorithm.Based on the outputs of steps 1 through 3, the induced scheduling problem is clas- sified and one or more algorithmic approaches are identified. The problem class may correspond to general allocation, multidimensional bin-packing, job-shop scheduling, or minimum-cost flow, depend- ing on the structur...

[5] [5]

Define the implementation and data model. The abstract model components are translated into concrete resource and workload descriptors, schemas, and platform APIs, mapping abstract re- quirements unambiguously onto the computing re- sources and domain constraints of the target envi- ronment. Extensions such as finer-grained configura- tion, compliance met...

[6] [6]

The first is to surface cross- layer interactions that are not apparent when layers are treated as independent components

Validate, iterate, and evolve.Validation pur- sues two distinct aims. The first is to surface cross- layer interactions that are not apparent when layers are treated as independent components. The second is to resolve conflicts between the abstract model and practical implementation constraints: platform or domain limitations discovered during implementa-...

[7] [7]

R. Fang, R. Bindu, A. Gupta, and D. Kang, LLM agents can autonomously exploit one-day vulnerabilities (2024), arXiv:2404.08144 [cs.CR]

arXiv 2024

[8] [8]

Anthropic,Disrupting the First Reported AI-Orchestrated Cyber Espionage Campaign, Tech. Rep. (Anthropic, 2025)

2025