{"id":"66f65496-7f93-4aff-b802-d56140a9f31a","arxiv_id":"2504.15198","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"DESTinE is a SimPy-based simulator that achieves fast large-scale cyber-network simulation for power grids and uses it to rank critical nodes and propose a hybrid topology.","lead":"This paper introduces DESTinE, a SimPy-based simulator that models communication networks overlaid on large synthetic power grids with 500 to 10,000 buses. It reports faster-than-real-time runtimes, ranks critical nodes under denial-of-service attacks, and proposes a hybrid star-radial topology to improve grid resilience.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Hybrid topology's 'improved resilience' is never tested under DoS: Table IV reports only simulation wall-clock times.","rationale":"I read the paper as a tool demonstration whose core scalability claim, faster-than-real-time standalone simulation of the 10,000-bus case (1.45 s, Table IV), is credible and internally consistent. The CORE comparison on a five-router topology provides a limited but real independent anchor, although only relative ordering is checked. The load-bearing weakness is not the simplified network abstraction per se (the authors disclose it in Remark 1), but the fact that the highest-level claim, the hybrid topology's resilience improvement, is never subjected to the adversarial scenario that motivated it. The optimizer output is reused to select the hybrid, and then only wall-clock completion time is reported, creating a circular validation for resilience. The reader's weakest-assumption pointed to the abstraction; I see the missing hybrid DoS evaluation as the sharper, more decisive gap. This does not move the verdict: CONDITIONAL remains appropriate because the scalability claims are supportable while the resilience claim requires the proposed check.","tokens_in":26500,"tokens_out":3484,"duration_ms":32622,"concrete_test":"Re-run the Section IV-B DoS experiment on the hybrid topology for ACTIVSg2000 and ACTIVSg10k: set each utility router's forwarding probability to 0.01, keep identical packet generation parameters and simulation duration, and record at the regulatory sinks the average end-to-end delay, packet loss ratio, and fraction of substations whose packets fail to arrive, comparing hybrid against star and radial. If the hybrid does not dominate both constituent topologies on at least delay and packet loss under DoS, the abstract's 'improve network resilience' claim is not supported. Report means over multiple seeds.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section IV-E derives the hybrid topology from severity classifications assigned by the DESTinE optimizer (Eq. 7), and the abstract claims it 'combines the strengths of star and radial structures to improve network resilience.' However, the only quantitative evaluation of the hybrid is Table IV in Section IV-F, which lists DESTinE's simulation completion time (seconds) under three configurations. Completion time is a computational scalability metric, not a resilience metric: it says nothing about end-to-end delay, packet loss, or delivery ratio under the DoS conditions used to identify critical nodes. The DoS attack (forwarding probability 0.01) is never re-run on the hybrid topology, so there is no evidence the hybrid actually reduces the impact of node compromise. Moreover, the hybrid's Table IV entries are not consistently faster (on the Raspberry Pi 5, hybrid equals star for ACTIVSg10k and is slower than radial for ACTIVSg2000), so even the efficiency interpretation is configuration-dependent. Because the hybrid topology is a headline contribution and the paper's motivation is grid resilience, this missing adversarial comparison is the most load-bearing gap.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents DESTinE, a SimPy-based discrete-event simulator for the cyber layer of large-scale power systems. It overlays star and radial communication topologies on the ACTIVSg500, ACTIVSg2000, and ACTIVSg10k synthetic grids, ranks utility routers using centrality metrics and simulation-derived delays under normal and DoS conditions, introduces a constrained optimization to classify utilities into six severity levels, and proposes a hybrid star/radial topology selected per substation based on those classifications. Performance is reported in standalone mode, with a virtual server, and with a Raspberry Pi 5 hardware-in-the-loop setup, and results are compared with the CORE emulator. The paper claims that DESTinE is efficient and scalable for large-scale cases and that the hybrid topology improves network resilience.","tokens_in":26773,"tokens_out":8348,"duration_ms":74007,"significance":"If the claims were fully substantiated, DESTinE would be a useful open-science contribution: it demonstrates faster-than-real-time simulation of the cyber layer for synthetic grids with up to 10,000 buses, integrates with emulation and hardware, and ships a dataset link for reproducibility. The scale alone is a strength relative to many existing cyber-power co-simulation tools, and the comparison with CORE, though small, is a reasonable first step. However, the headline resilience claim for the hybrid topology is not actually tested, and the ranking/optimization pipeline is largely self-referential. The paper's value for practitioners depends on closing those gaps; with additional validation the tool and the methodology could be a solid contribution to scalable cyber-physical security assessment.","major_comments":[{"comment":"The central claim that the hybrid topology 'improve[s] network resilience' is not supported by the reported experiments. Table IV reports only simulation wall-clock completion times for the three configurations. No DoS attack is re-run on the hybrid topology, and no end-to-end delay, packet-loss, or delivery-ratio metrics are given for any topology. In fact, the hybrid is slower than radial for ACTIVSg2000 and equal to star for ACTIVSg10k on the Raspberry Pi 5 configurations, so even the computational-efficiency interpretation is configuration-dependent. A resilience evaluation must directly compare star, radial, and hybrid under the same DoS scenario using network performance metrics, not just simulation runtime.","section":"Abstract; Section IV-F, Table IV"},{"comment":"The ranking and topology-selection pipeline is self-referential. Equation (5) constrains |S(v)-T(v)| <= 1, and Eq. (6) enforces |S-T|^2 = 0 or 1, so the optimized ranking T(v) is essentially the simulation ranking S(v) (or S(v) plus or minus one rank). The same T(v) is then used to define the severity classes (via the alpha0/alpha1 values) and, through Eq. (7), to select the hybrid topology. Moreover, the reported weights a=0.97, b=0.02, c=0.01 give the largest objective weight to closeness centrality (1/c = 100), despite the text stating that the simulation ranks correlate most strongly with betweenness centrality. As a result, the optimization does not provide an independent, topology-informed validation of the simulation-based criticality ranking. I recommend evaluating T(v) on held-out attack scenarios or comparing it against an external ground truth.","section":"Section III-C, Eq. (5); Section IV-D, Eq. (6) and Section IV-E, Eq. (7)"},{"comment":"The fidelity of the delay and congestion model is not established. Packet sizes are exponentially distributed with a mean of 3.4 MB and routers serve at 2.2 packets/sec, which corresponds to roughly 60 Mbps and is far outside typical SCADA/IEC 61850 traffic profiles. Remark 1 explicitly excludes Layer 2/1 overhead, routing protocol overhead, and device processing delays. The CORE comparison in Section IV-C uses only five routers, 100-byte packets, and matches only the relative order of delays, not absolute values. Without a sensitivity analysis over traffic parameters or validation against a realistic traffic profile, the critical-node rankings and severity classes derived from S(v) are not tied to real network behavior.","section":"Section IV-B, Table II, Remark 1; Section IV-C"},{"comment":"The optimization formulation is internally inconsistent as written. The two constraints |S(v)-T(v)|^2 = 0 (with alpha = alpha0) and |S(v)-T(v)|^2 = 1 (with alpha = alpha1) cannot hold simultaneously for a single decision variable T(v); the text should state that these are two separate optimization runs. In addition, the problem is described as convex, but T(v) is integer-valued (0 to k), making it a mixed-integer problem; the convexity claim needs clarification. Because the resulting alpha0 and alpha1 values drive the entire severity classification and the hybrid topology selection, this ambiguity affects the reproducibility of the paper's main results.","section":"Section IV-D, Eq. (6)"},{"comment":"The scalability evidence lacks statistical grounding. All completion-time entries in Table IV appear to be single measurements, with no repeated trials or error bars, despite the large variability of the Raspberry Pi 5 configurations (e.g., ACTIVSg500 at 17.00 s versus ACTIVSg10k at 208.00 s). At minimum, multiple runs with mean and standard deviation should be reported, particularly for the hardware-in-the-loop setup where WiFi and system load conditions fluctuate.","section":"Section IV-F, Table IV"}],"minor_comments":[{"comment":"The section header misspells ACTIVSg2000 as 'ACTiVSG2000'.","section":"Section IV-E"},{"comment":"The table labels 'Max = 0.05 sec' and 'Max = 20 samples/sec' for exponential distributions; an exponential distribution has no maximum, so please clarify whether these are means, upper bounds on sampled values, or parameters of a truncated exponential.","section":"Table II"},{"comment":"Closeness centrality is denoted inconsistently as Cc(v) in the objectives and CC(v) in Eq. (4); unify the notation.","section":"Equations (5) and (6)"},{"comment":"Typo: 'cyber-phyiscal' should be 'cyber-physical'.","section":"Section I-A"},{"comment":"The statement that 'DESTinE successfully simulated over 4700 nodes in approximately 200 seconds with HIL integration' is not directly identifiable in Table IV; please state which row and configuration this refers to.","section":"Section IV-C"},{"comment":"The text says 'each router could generate at least one packet but no more than two packets,' while Table IV lists exact packet counts; please specify how these counts are determined from the topology.","section":"Section IV-F"},{"comment":"The phrase 'allowing the reviewer to further analyze the cases in detail' should be 'allowing readers to analyze the cases in detail.'","section":"Appendix / Dataset"}],"recommendation":"major_revision","confidential_remarks":"The paper is borderline between major revision and reject because the central hybrid-resilience claim is headline material and is currently unsupported. However, the simulator itself appears functional, the scale is impressive, and the missing experiments and formulation clarifications are within the scope of a revision. The optimization circularity is a deeper concern but can be addressed by repositioning the optimization as a ranking aid rather than an independent validation. I would not reject if the revision adds a DoS evaluation of the hybrid topology and tightens the claims."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, DESTinE is a credible engineering artifact: SimPy-based discrete-event simulation over ACTIVSg500/2000/10k cyber overlays, with centrality-based critical-node ranking, severity classification, virtual-server and Raspberry Pi HIL integration. The standalone 10k-bus run finishing in 1.45 s is a genuinely useful screening capability. Second, the hybrid topology's 'improved resilience' claim is not actually tested. The hybrid is built from severity rankings that come essentially from the simulation itself, and the only quantitative evaluation in Table IV is simulation completion time. That is a scalability metric, not a resilience metric. The DoS attack that motivates the paper is never re-run on the hybrid, and no end-to-end delay, packet loss, or delivery ratio is reported for it. The abstract overstates what the evidence supports. The paper does several things well. It is transparent about abstraction limits in Remark 1, it provides detailed appendix tables of rankings and severity classes, it makes a small CORE comparison, and it offers a dataset link. The scalability story is plausible and useful for practitioners who want to screen large cyber-physical systems for weak nodes. The soft spots are real but fixable. Beyond the missing hybrid validation, the optimization in Eq. (6) constrains T(v) to stay within one rank of the simulation output S(v), so the 'optimized' ranking is largely a re-encoding of the simulation. That is not a fatal flaw if the simulation is the ground truth, but it means the severity classes and hybrid selection inherit all the biases of the simplified traffic model. Timings also lack repeated trials and error bars, and the CORE comparison matches only relative delay order on five routers. These are minor-to-moderate weaknesses, not load-bearing errors. The central scalability result stands: the tool runs fast and the paper is readable. Who is this for? Utility planners and researchers who want a fast cyber-layer screening tool for large synthetic grids, especially those already using ACTIVSg test cases. With a proper DoS comparison on the hybrid and repeated timing trials, this would be a solid contribution. As is, it deserves a serious referee but needs revision. My recommendation: send it to peer review with explicit requests for a DoS-based resilience evaluation of the hybrid and a clearer separation between efficiency claims and resilience claims.","headline":"A useful, fast cyber-layer simulator for large synthetic grids, but the paper's headline hybrid-topology resilience claim is unsupported: no DoS test is re-run on the hybrid and Table IV only reports wall-clock times.","tokens_in":27266,"tokens_out":2087,"would_cite":false,"duration_ms":22161,"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":"DESTinE ranks critical utility nodes under denial-of-service attacks faster than real time and derives a hybrid star-radial topology that improves network resilience.","keywords":["discrete event simulation","cyber-physical power systems","denial of service","network centrality","hybrid topology","scalability","hardware-in-the-loop","synthetic power grids"],"falsifier":"Run the same star and radial overlays on the ACTIVSg2000 and ACTIVSg10k cases in a high-fidelity emulator with the identical traffic and denial-of-service conditions and compare per-utility average-delay rankings; if DESTinE's ranking of the 20 or 80 utilities differs from the emulator's by more than one rank for more than a small fraction of utilities, the severity classification and hybrid topology would not transfer. A cheaper check is absolute delay: DESTinE excludes Layer 2/1 and processing overhead, so its delays should be systematically lower than a full-stack emulator, and any utility pair whose relative delay order flips would break the optimization's weighting of simulation ranks.","tokens_in":26319,"feed_emoji":"⚡","tokens_out":10932,"duration_ms":85914,"temperature":0.7,"pith_summary":"This paper presents DESTinE, a discrete event simulation tool for the cyber layer of large-scale power systems. DESTinE generates and routes network packets faster than real time—simulating the 10,000-bus ACTIVSg10k case in 1.45 seconds standalone—while modeling congestion and packet loss under normal and denial-of-service conditions. By combining simulation-based delay rankings with graph centrality measures through a constrained optimization, DESTinE classifies utility nodes into six severity levels and selects, for each substation, the star or radial connection that minimizes risk, producing a hybrid topology. The intended payoff is a scalable way for grid operators and planners to assess cyber risk, prioritize defenses, and reconfigure communication networks in near real time.","feed_headline":"Simulator ranks grid cyber risk on 10,000 buses in 1.45 seconds","feed_subtitle":"Faster-than-real-time packet simulation ranks critical utility nodes and builds a resilient hybrid topology.","key_machinery":"The load-bearing mechanism is the DESTinE event-driven simulation loop built on SimPy's generator-based processes: five device classes—Packet Generator, Router, Port Monitor, Sink, and Random Brancher—create, forward, queue, monitor, and consume packets, with routers dropping packets when queue capacity is exceeded and forwarding at an average 2.2 packets/sec. Two further pieces do the analytical work: the network analysis matrices (global: the normalized-Laplacian spectral gap and the global clustering coefficient; local: betweenness, eigenvector, and closeness centrality) and the constrained optimization of Equation (6), which reconciles simulation ranks with centrality ranks under a convex objective with weights a,b,c and penalty parameters α0 and α1. The hybrid topology follows from the per-substation rule Th(s), which selects star when its severity score is lower than radial and radial otherwise, with radial as the tie-breaker.","core_discovery":"The central claim is that DESTinE—built on SimPy's process-driven event simulation with Packet Generator, Router, Port Monitor, Sink, and Random Brancher components—reproduces the delay and congestion behavior of large communication networks well enough to rank cyber-node criticality, and does so at scales emulators cannot reach. On the ACTIVSg500, ACTIVSg2000, and ACTIVSg10k synthetic grids with star and radial cyber overlays, a DoS attack is modeled by dropping each target router's forwarding probability to 0.01; the resulting average-delay rankings of the 4, 20, and 80 utility routers are fused with betweenness, eigenvector, and closeness centrality via a convex optimization whose weights settle at a=0.97, b=0.02, and c=0.01. The optimizer's α0 and α1 penalty parameters, together with the number of substations lost, place each utility into one of six severity classes, and a per-substation rule Th(s) picks star or radial connection to minimize risk, yielding a hybrid topology that the paper reports achieves the shortest simulation completion times in standalone and virtual-server runs. The paper also reports that the relative delay order agrees with the CORE emulator on a five-router test network, and demonstrates integration with a Node.js virtual server and a Raspberry Pi 5 over WiFi 6.","pith_inferences":["If DESTinE's delay-order fidelity holds, the same optimization pipeline could be reused for other attack classes—false data injection, man-in-the-middle, and replay—by altering packet content or routing behavior in the event loop; the paper describes these templates but does not test them.","The hybrid selection rule Th(s) is a greedy per-substation choice; a natural extension is to solve a global assignment over all substations and regulatory routers simultaneously, which might beat the greedy rule in worst-case attacks.","The near-zero spectral gap of every tested graph indicates these networks are highly partitionable; that suggests defender placement at partition boundaries could protect against node loss at least as effectively as reconfiguring to a hybrid topology—a comparison the paper does not run.","A decisive extension would be to validate absolute delay and loss magnitudes against a full protocol-stack emulator or live SCADA traffic; the paper's CORE comparison establishes only relative ordering on a five-router network."],"forward_implications":["DESTinE simulates the cyber layer of the 10,000-bus system in about 1.45 seconds standalone, so cyber-layer analysis at the scale of continental synthetic grids becomes practical in near real time (Table IV).","The optimizer's weights place most of the ranking weight on betweenness centrality (a=0.97), so the paper's critical-node ordering is primarily a structural property of the communication graph rather than a detail of the traffic model.","The hybrid topology, built by choosing star or radial per substation based on lower severity, matches or beats the pure topologies in simulation completion time in standalone and virtual-server configurations, indicating the reconfiguration does not sacrifice speed.","Integrating DESTinE with a Node.js virtual server and a Raspberry Pi 5 allows real TCP packets to traverse a WiFi link while the simulator still completes the 10,000-bus case in about 208 seconds, enabling hardware-in-the-loop cyber testing at scale.","Because all tested graphs are sparse with spectral gaps at or near zero, the paper argues the optimization converges quickly and the framework transfers to other sparse power-system communication topologies."],"supporting_citations":[{"why":"supplies the scalable automatic model generation that produces the JSON substation/utility/regulatory router models DESTinE simulates","marker":"[11]"},{"why":"establishes the comparison of graph-theory metrics with time-domain discrete-event simulation that the optimization fuses","marker":"[12]"},{"why":"SimPy, the Python discrete-event engine that DESTinE is built on","marker":"[13]"},{"why":"CORE emulator used as the comparison baseline for relative delay ordering","marker":"[17]"},{"why":"ACTIVSg500 synthetic grid test case","marker":"[30]"},{"why":"ACTIVSg2000 synthetic grid test case","marker":"[31]"},{"why":"ACTIVSg10k synthetic grid test case with its regulatory-unit jurisdictions","marker":"[32]"},{"why":"defines the normalized-Laplacian spectral gap used in the global matrix analysis","marker":"[39]"},{"why":"supplies the N-1 contingency/SCOPF parallel that frames the DoS-based network contingency optimization","marker":"[44]"}],"fun_headline_variants":["DESTinE ranks grid cyber risk at 10,000-bus scale","Faster-than-real-time grid cyber simulation scales to 10k buses","Simulator identifies critical nodes on 10,000-bus power grids","DESTinE scales cyber-physical simulation to 10k-bus networks","Tool ranks grid cyber weak spots in faster-than-real-time"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the simplified network model—exponential packet sizes averaging 3.4 MB, router service at 2.2 packets per second, and denial of service as a forwarding probability of 0.01, with Layer 2/1 overhead and device processing delays excluded—preserves the relative order of delays and packet losses of real SCADA networks; if that order shifts, the critical-node rankings, the six severity classes, and the hybrid topology lose their empirical basis.","fun_headline_variants_meta":{"raw":{"variants":["DESTinE ranks grid cyber risk at 10,000-bus scale","Faster-than-real-time grid cyber simulation scales to 10k buses","Simulator identifies critical nodes on 10,000-bus power grids","DESTinE scales cyber-physical simulation to 10k-bus networks","Tool ranks grid cyber weak spots in faster-than-real-time"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000192,"raw_usage":{"total_tokens":1413,"prompt_tokens":1079,"completion_tokens":334,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":695,"completion_tokens_details":{"reasoning_tokens":237}},"tokens_in":695,"tokens_out":334,"duration_ms":3620,"temperature":1.0,"reasoning_tokens":237,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T11:31:12.996839+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the same star and radial overlays on the ACTIVSg2000 and ACTIVSg10k cases in a high-fidelity emulator with the identical traffic and denial-of-service conditions and compare per-utility average-delay rankings; if DESTinE's ranking of the 20 or 80 utilities differs from the emulator's by more than one rank for more than a small fraction of utilities, the severity classification and hybrid topology would not transfer. A cheaper check is absolute delay: DESTinE excludes Layer 2/1 and processing overhead, so its delays should be systematically lower than a full-stack emulator, and any utility pair whose relative delay order flips would break the optimization's weighting of simulation ranks.","supporting_citations":[{"cited_title":"A scalable automatic model generation tool for cyber-physical network topologies and data flows for large-scale synthetic power grid models,","cited_arxiv_id":null,"evidence_quote":"supplies the scalable automatic model generation that produces the JSON substation/utility/regulatory router models DESTinE simulates"},{"cited_title":"On graph theory vs. time-domain discrete-event simulation for topology-informed assessment of power grid cyber risk,","cited_arxiv_id":null,"evidence_quote":"establishes the comparison of graph-theory metrics with time-domain discrete-event simulation that the optimization fuses"},{"cited_title":"Simpy: Discrete event simulation for python,","cited_arxiv_id":null,"evidence_quote":"SimPy, the Python discrete-event engine that DESTinE is built on"},{"cited_title":"Performance analysis of 5g network slicing simulations using simpy,","cited_arxiv_id":null,"evidence_quote":"CORE emulator used as the comparison baseline for relative delay ordering"},{"cited_title":"A metric-based validation process to assess the realism of synthetic power grids,","cited_arxiv_id":null,"evidence_quote":"ACTIVSg500 synthetic grid test case"},{"cited_title":"Design of next-generation cyber-physical energy manage- ment systems: Monitoring to mitigation,","cited_arxiv_id":null,"evidence_quote":"ACTIVSg2000 synthetic grid test case"},{"cited_title":"Power flow convergence and reactive power planning in the creation of large synthetic grids,","cited_arxiv_id":null,"evidence_quote":"ACTIVSg10k synthetic grid test case with its regulatory-unit jurisdictions"},{"cited_title":"Wireshark user’s guide,","cited_arxiv_id":null,"evidence_quote":"defines the normalized-Laplacian spectral gap used in the global matrix analysis"}],"review_version":1}