{"id":"a8a9a86c-b34a-476d-ad03-274fa8c64a01","arxiv_id":"2504.20389","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A multi-tenant distributed quantum computing framework combining circuit placement with priority-based network scheduling cuts simulated job completion time versus earlier single-circuit methods.","lead":"CloudQC is a framework for placing multiple quantum circuits onto a network of quantum processors and scheduling the entangled connections between them. It is an early attempt to run distributed quantum computing as a multi-tenant cloud, and simulations on benchmark circuits report large reductions in job completion time.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline JCT-vs-existing-algorithms claim is not directly tested: placement experiments measure communication cost, and JCT runs compare only CloudQC ablations or simple schedulers, not prior DQC placement algorithms.","rationale":"The reader's weakest assumption targets the missing definitions of p_u(x_u) and d_u(x_u) in Eq. 10 and Algorithm 3. That is a legitimate reproducibility and transferability concern: the network scheduler's benefit depends on how redundant communication-qubit allocations translate into success probability and duration, and the paper never specifies these functions. I agree that this gap matters for whether the simulated JCT reductions would transfer to real hardware. However, the most load-bearing problem for the central claim is upstream: the claimed comparison against existing DQC placement algorithms on job completion time never appears. A paper that announces a significant JCT reduction over existing algorithms must include that comparison; instead, single-circuit placement is scored only by remote-operation count, and the JCT evaluations compare schedulers or internal ablations. The unstated p_u/d_u model affects predictive realism, but the missing baseline affects whether the abstract's comparative statement is evidenced at all. The public code makes the required check easy, and the missing comparison is fixable. I would therefore keep the conditional verdict rather than reject, while asking for the end-to-end baseline comparison and a precise statement of the scheduler's resource-to-latency model.","tokens_in":16462,"tokens_out":5631,"duration_ms":60878,"concrete_test":"Use the public CloudQC simulator to rerun the Section VI workloads (e.g., mixed, QFT, arithmetic, and the single-circuit benchmarks) under the same 20-QPU topology settings, but replace the placement step with SA [11] and GA [38] plus Random, keeping the identical network scheduler for all placements. Measure average and CDF of job completion time over the same number of batches and topologies. If CloudQC's JCT advantage over at least one existing placement baseline is not reproducible or statistically clear, the headline claim should be narrowed to placement cost and scheduling improvements under CloudQC's own model.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and conclusion claim that CloudQC 'significantly reduces the average job completion time compared to existing DQC placement algorithms for both single-circuit and multi-circuit DQC.' The reported experiments do not actually compare job completion time against an existing DQC placement algorithm. Table III evaluates placement only by number of remote operations, not completion time, against SA, GA, and Random. The JCT results in Section VI-C (Fig. 22) compare Greedy, Average, and Random resource allocation while holding the placement fixed, so they isolate the scheduler rather than the full framework. The multi-tenant results in Section VI-D compare only CloudQC-BFS and CloudQC-FIFO, both ablations of the proposed framework rather than prior DQC placement algorithms. Since the central claim is an end-to-end performance comparison, the missing experiment is load-bearing: even if the unstated p_u(x_u) and d_u(x_u) model in Eq. 10 is realistic, the paper does not show that the complete CloudQC pipeline beats an existing placement algorithm on the metric claimed. This is an evidence gap rather than a demonstrated error, and it is addressable by running the public simulator with the prior baselines.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces CloudQC, a framework for multi-tenant distributed quantum computing (DQC) that combines circuit placement and quantum network scheduling. The placement component partitions quantum circuits, selects QPUs via modularity-based community detection, and maps partitions using a graph-center heuristic; the network scheduler allocates communication qubits to remote operations based on a critical-path-inspired priority. The authors claim that CloudQC is the first framework to address placement and network scheduling for multiple concurrent DQC circuits, and that simulations with real quantum circuit workloads show it significantly reduces job completion time compared to existing DQC placement algorithms for both single-circuit and multi-circuit DQC. A public simulator is provided. The paper includes an NP-hardness argument for the placement problem and evaluations of placement cost, network scheduling, and multi-tenant scenarios.","tokens_in":16768,"tokens_out":4485,"duration_ms":43917,"significance":"If the central claim is substantiated, the paper addresses a genuine gap: prior DQC placement work focuses on single circuits, and classical cloud placement techniques do not account for the probabilistic nature of quantum communication. The public simulator, use of QASMBench workloads, and incorporation of EPR success probabilities are concrete strengths that support reproducibility and further research. The network scheduler's redundancy idea is interesting and potentially useful. However, the evidence presented does not yet establish the headline claim that job completion time is significantly reduced compared to existing DQC placement algorithms, because the reported experiments do not directly compare end-to-end placement-and-scheduling performance against prior placement methods. The contribution is a plausible stepping stone, but the evaluation needs to be strengthened before the claim is supportable.","major_comments":[{"comment":"The central claim, stated in the abstract and conclusion, is that CloudQC 'significantly reduces the average job completion time compared to existing DQC placement algorithms for both single-circuit and multi-circuit DQC.' This claim is not directly tested by the reported experiments. Table III in Section VI-B evaluates placement only by the number of remote operations (communication cost), not by job completion time, against SA, GA, and Random. Section VI-C (Fig. 22 and related figures) compares CloudQC with Greedy, Average, and Random resource allocation while holding the placement fixed, which isolates the network scheduler rather than the full placement-plus-scheduling pipeline. Section VI-D compares only CloudQC-BFS and CloudQC-FIFO, both ablations of the proposed framework, against CloudQC in multi-tenant settings; no prior DQC placement algorithm is included. Consequently, the end-to-end JCT comparison against existing DQC placement algorithms is missing. This is a load-bearing evidence gap that can be addressed by running the public simulator with SA, GA, and Random as full end-to-end baselines and reporting JCT for both single-circuit and multi-circuit workloads.","section":"Abstract and Section VI"},{"comment":"The network scheduling benefit rests on an unspecified model of how allocating more communication qubits x_u reduces the expected duration of a remote operation. Eq. (10) defines completion time as c_u = s_u + d_u(x_u), and the text refers to success probability p_u(x_u), but neither d_u(x_u) nor p_u(x_u) is ever defined. Algorithm 3 contains steps such as 'Identify competing sets for resource allocation,' 'Allocate resources between competing sets,' and 'Determine resources allocation based on the priority of node' without specifying the actual allocation rules. Without these definitions, the network scheduling results cannot be reproduced or transferred to hardware, and the reported JCT reductions may be an artifact of an assumed (and unstated) model of how redundancy improves remote-gate latency. The authors should state the functional forms of d_u(x_u) and p_u(x_u), or provide the scheduling equations that Algorithm 3 implements.","section":"Section IV-C, Eq. (10), Algorithm 3"},{"comment":"The placement scoring function S = α × (1/T) + β × (1/C) and the batch manager metric I_i = λ1·(#CNOTs/n_i) + λ2·n_i + λ3·d_i both contain weights (α, β, λ1, λ2, λ3) that are never specified in the paper. No sensitivity analysis is reported. Since these weights directly influence which placement and job order are chosen, the reported improvements could depend on a particular choice of weights. The authors should report the values used in the experiments and show that the conclusions are robust across a reasonable range of weight settings.","section":"Section V-B, scoring function and Eq. (11)"},{"comment":"The performance comparisons lack error bars, confidence intervals, or statistical significance tests. Section VI-D states that each batch is run 20 times with different network topologies, but only CDF curves are shown without any uncertainty quantification. The text repeatedly uses 'significantly' (e.g., 'CloudQC significantly achieves the least job completion time' in Section VI-C.1) without supporting statistical analysis. At minimum, the authors should report the variance across random topologies and EPR-generation trials and, where appropriate, perform paired significance tests against the baselines.","section":"Section VI, all subsections"}],"minor_comments":[{"comment":"There are several typos and naming inconsistencies: 'CloudDC' appears twice in Section V-C instead of 'CloudQC'; Section VI-C contains 'flowing scheduling' instead of 'flow scheduling'; the section heading VI-D uses 'Multi-Tennant' instead of 'Multi-Tenant'; and Section IV-C has the fragment 'we With this example'. The example about gate importance refers to 'the first gate' and then 'the ninth gate' in an unclear way; the text should consistently identify the gates being compared.","section":"Section IV-C and V-C"},{"comment":"The column header 'CdQC-BFS' appears to be a typo for 'CloudQC-BFS'. Additionally, the caption says 'NUMBER OF REMOTE OPERATIONS OF SINGLE-CIRCUIT PLACEMENT' but the abstract and text emphasize job completion time; the table should clarify that this metric is communication cost only, not JCT.","section":"Table III"},{"comment":"The NP-hardness argument for the multi-tenant problem is sketched with 'Due to the page limitation, we omit the details here.' Since NP-hardness of the multi-tenant placement is claimed as a contribution, the reduction should be given in an appendix or stated more carefully; otherwise, the claim remains unsupported.","section":"Section IV-B"},{"comment":"The y-axis is labeled 'Relative Job Completion Time' but the text does not state what the values are relative to. The caption should specify the normalization baseline (presumably CloudQC's completion time is set to 1, or the minimum across methods).","section":"Figure 22"}],"recommendation":"major_revision","confidential_remarks":"The paper addresses a timely and relevant problem, and the public simulator is a valuable asset for the community. The main issue is that the abstract and conclusions overstate what the experiments show: JCT is never compared end-to-end against an existing DQC placement algorithm. The authors can fix this by adding those baseline experiments using their own simulator, which they explicitly say is public. If the added experiments support the current claims, the paper would be a solid systems contribution. I would not reject: the gap is in the evaluation, not in the core approach, and it is addressable within the manuscript's scope."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the good news: CloudQC is the first thing I've seen that treats multi-tenant distributed quantum computing as a combined placement-and-network-scheduling problem, and that framing is the real contribution. Single-circuit DQC placement and single-QPU multi-programming are well-trodden; the multi-tenant DQC setting is not. The paper also ships code, uses QASMBench circuits, and borrows latency/EPR parameters from real experiments, which makes the evaluation reproducible in principle.\n\nThe placement evaluation is genuinely encouraging. On remote-operation count, CloudQC beats SA, GA, and Random by wide margins on most circuits, often 5-20x. The network scheduler also beats Greedy, Average, and Random on JCT for structured circuits like QFT and QV. Those results support the design choices.\n\nNow the soft spots, and the reader/stress-test have them right. The abstract claims CloudQC 'significantly reduces average job completion time compared to existing DQC placement algorithms'—but no experiment actually measures JCT against an existing DQC placement algorithm. Table III compares single-circuit placement on remote operations, not JCT. The JCT experiments in Section VI-C hold placement fixed and compare scheduler policies; the multi-tenant results in VI-D compare only CloudQC's own ablations (BFS, FIFO). To test the central claim, the authors need to run their simulator with prior baselines like Autocomm, Qucomm, or Mao et al.'s SA on the JCT metric. That's a missing experiment, not a demonstrated error, and it's addressable because the code is public.\n\nSecond, the network scheduler's model of how extra communication qubits reduce remote-gate duration is never written down. Eq. 10 has d_u(x_u), and the whole redundancy argument depends on it, but no definition or formula appears. That makes the JCT gains hard to interpret. The placement scoring weights α, β and batch-manager weights λ1-3 are also unspecified, with no sensitivity analysis. These are minor-to-moderate in isolation, but combined with the missing baseline they undercut confidence in the headline numbers.\n\nOn circularity: the evaluation uses the authors' own simulator, but the latency and EPR success values come from external experiments, and the hyperparameters aren't fitted to the metric. That's acceptable for a system paper, though error bars on the single-circuit tables would help.\n\nBottom line: the problem is real, the framework is coherent, and the evidence is promising but incomplete. This deserves a serious peer review, conditional on the authors adding a real JCT comparison against prior placement algorithms and stating the resource-latency model.","headline":"First framing of multi-tenant DQC placement and scheduling, with a real evaluation gap: the headline JCT claim is not tested against prior DQC placement algorithms.","tokens_in":17225,"tokens_out":2681,"would_cite":true,"duration_ms":25968,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"CloudQC is the first framework for placing and network-scheduling multiple concurrent distributed quantum circuits across a multi-tenant quantum cloud, and its simulations show significant job completion time reductions.","keywords":["distributed quantum computing","quantum cloud","circuit placement","network scheduling","EPR pairs","multi-tenant","job completion time","remote gate DAG"],"falsifier":"A testbed or hardware-accurate simulation that measures remote-gate latency as a function of the number of communication qubits allocated to EPR generation attempts. If expected latency is nearly flat beyond one or two attempts, or if EPR attempts on different qubits interfere, then CloudQC's priority-based allocation would produce no job-completion-time benefit, and its reported reductions would vanish.","tokens_in":16287,"feed_emoji":"⚛️","tokens_out":4512,"duration_ms":40582,"temperature":0.7,"pith_summary":"CloudQC is the first framework that tackles both placement and network scheduling for multiple distributed quantum computing (DQC) circuits running concurrently in a shared quantum cloud. It argues that prior DQC placement methods handle only a single circuit and ignore the probabilistic nature of EPR-pair generation, so they cannot work in a multi-tenant setting. CloudQC combines graph-partition-based circuit placement, community detection to choose QPU groups, and a network scheduler that gives extra communication qubits to remote gates that lie on longer critical paths. Simulations with real quantum circuit workloads show that CloudQC reduces average job completion time compared to existing DQC placement algorithms for both single-circuit and multi-circuit DQC.","feed_headline":"CloudQC cuts DQC job time with smarter EPR allocation","feed_subtitle":"First placement and scheduling framework for many concurrent quantum circuits; simulations on real workloads show clear gains.","key_machinery":"The remote-gate DAG (directed acyclic graph) of each placed circuit, where nodes are inter-QPU two-qubit gates and edges are their logical dependencies, carries the network scheduler's argument. Each gate receives a priority equal to the length of the longest path from it to a leaf node; the scheduler then allocates more communication qubits (hence more EPR generation attempts) to higher-priority gates, because failure of those gates would backlog more of the circuit. The placement component uses a modularity-based community detection algorithm to select QPU groups with strong connectivity and sufficient computing qubits, and a scoring function over estimated running time and communication cost to pick the final placement.","core_discovery":"On its own terms, the paper claims that placing multiple quantum circuits in a shared QPU cluster requires joint optimization of circuit placement and network resource allocation, and that a scheduler exploiting the remote-gate DAG's structure can significantly cut job completion time. The central discovery is a concrete framework: partition each circuit, map partitions to QPUs using community detection and a center-based heuristic, then allocate communication qubits per remote gate in proportion to the gate's priority, defined as the longest path length from that gate to a leaf in the remote DAG. The paper reports that this priority-based redundant allocation outperforms greedy, average, and random allocation strategies on real circuit workloads, and that the placement step alone reduces communication overhead relative to simulated annealing, genetic algorithm, and BFS-based placement.","pith_inferences":["The priority metric (longest path to a leaf) is a crude critical-path proxy; a scheduler that weights paths by EPR success probability and gate execution times would likely do even better, but the paper does not test this.","The framework assumes a fixed topology and homogeneous QPU capabilities; extending it to heterogeneous QPUs with different error rates or link reliabilities is a natural next step that the placement score could encode via edge weights.","The unstated latency model is the main risk: if redundant EPR attempts do not reduce expected remote-gate duration on real hardware, the framework's scheduler advantage could shrink, while its placement component may still stand."],"forward_implications":["If CloudQC holds, future quantum cloud platforms should treat circuit placement and network scheduling as one joint problem, not separate concerns.","Remote gates on long DAG paths are the right targets for redundant communication resources, so DQC compilers should compute path-length priorities before allocating EPR pairs.","The placement method's community-detection step gives a practical way to keep highly interacting partitions on nearby QPUs while preserving resources for future jobs.","Multi-tenant quantum clouds can improve throughput without changing underlying hardware, by scheduling existing EPR resources more intelligently.","The results suggest that improving EPR success probability itself is a key hardware lever, since all schedulers' completion times drop as that probability rises."],"supporting_citations":[{"why":"Supplies the simulated-annealing qubit allocation baseline for DQC that CloudQC compares against and improves.","marker":"[11]"},{"why":"Prior single-circuit DQC communication optimizer that motivates considering probabilistic communication in the cloud setting.","marker":"[7]"},{"why":"Prior collective communication optimization for DQC that provides a baseline for the placement problem.","marker":"[9]"},{"why":"A multinode quantum network experiment that provides the EPR success probability and operation latency values used in simulation.","marker":"[16]"},{"why":"The modularity-based community detection algorithm used to select feasible QPU sets in circuit placement.","marker":"[33]"},{"why":"The graph partitioning method used to divide circuits into balanced subgraphs while minimizing cut edges.","marker":"[34]"},{"why":"A benchmark suite of real quantum circuits used as the workloads in the evaluation.","marker":"[35]"},{"why":"Prior work on concurrent entanglement routing that motivates the EPR resource contention and success probability model.","marker":"[37]"}],"fun_headline_variants":["CloudQC: network-aware placement and scheduling for multi-tenant DQC","Priority-based EPR allocation in CloudQC cuts DQC job completion time","Multi-tenant DQC gets network-aware scheduling with CloudQC","CloudQC optimizes placement and EPR to shorten DQC jobs","CloudQC's network-aware scheduling reduces multi-tenant DQC job time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole network-scheduling gain rests on an unstated model in which allocating more communication qubits to a remote gate meaningfully reduces its expected duration through redundant EPR generation attempts; the paper never defines the functions $d_u(x_u)$ and $p_u(x_u)$ that encode this effect.","fun_headline_variants_meta":{"raw":{"variants":["CloudQC: network-aware placement and scheduling for multi-tenant DQC","Priority-based EPR allocation in CloudQC cuts DQC job completion time","Multi-tenant DQC gets network-aware scheduling with CloudQC","CloudQC optimizes placement and EPR to shorten DQC jobs","CloudQC's network-aware scheduling reduces multi-tenant DQC job time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001058,"raw_usage":{"total_tokens":4407,"prompt_tokens":883,"completion_tokens":3524,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":499,"completion_tokens_details":{"reasoning_tokens":3431}},"tokens_in":499,"tokens_out":3524,"duration_ms":24862,"temperature":1.0,"reasoning_tokens":3431,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T05:30:13.434739+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A testbed or hardware-accurate simulation that measures remote-gate latency as a function of the number of communication qubits allocated to EPR generation attempts. If expected latency is nearly flat beyond one or two attempts, or if EPR attempts on different qubits interfere, then CloudQC's priority-based allocation would produce no job-completion-time benefit, and its reported reductions would vanish.","supporting_citations":[{"cited_title":"Qucomm: Optimizing collective com- munication for distributed quantum computing,","cited_arxiv_id":null,"evidence_quote":"Prior collective communication optimization for DQC that provides a baseline for the placement problem."},{"cited_title":"Qubit allocation for distributed quantum computing,","cited_arxiv_id":null,"evidence_quote":"Supplies the simulated-annealing qubit allocation baseline for DQC that CloudQC compares against and improves."},{"cited_title":"Autocomm: A framework for enabling efficient communication in distributed quan- tum programs,","cited_arxiv_id":null,"evidence_quote":"Prior single-circuit DQC communication optimizer that motivates considering probabilistic communication in the cloud setting."},{"cited_title":"Modularity and community structure in networks,","cited_arxiv_id":null,"evidence_quote":"The modularity-based community detection algorithm used to select feasible QPU sets in circuit placement."},{"cited_title":"Qasmbench: A low- level quantum benchmark suite for nisq evaluation and simulation,","cited_arxiv_id":null,"evidence_quote":"A benchmark suite of real quantum circuits used as the workloads in the evaluation."},{"cited_title":"Concurrent entanglement routing for quantum networks: Model and designs,","cited_arxiv_id":null,"evidence_quote":"Prior work on concurrent entanglement routing that motivates the EPR resource contention and success probability model."}],"review_version":1}