{"id":"621850dd-f1eb-4c1d-9db7-0b1756113b3c","arxiv_id":"2607.20363","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"SRAN separates NDN name-prefix reachability from topological reachability, making core forwarding state grow with router topology, not application prefixes.","lead":"SRAN is a new routing scheme for Named Data Networking that maps application names to edge routers, so core routers forward on topology alone. It applies the 'map-and-encap' pattern from IP routing to NDN and adds BIER-based stateless multicast.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The headline claim that forwarding state scales with topology is false as stated: edge PETs grow linearly with prefix count, so total state is O(E·P), not topology-bound; Fig. 6's own data show edge state rising from 646 to 1726 entries.","rationale":"The reader's verdict is CONDITIONAL and notes the abstract overclaim as issue #3, but designates the PET miss path as the weakest assumption. In my reading, the miss path is a valid availability concern shared by any routing convergence process, and the paper's measured convergence times are comparable to baseline, so it is not the most load-bearing issue for the paper's central claim. Instead, the central claim — 'network forwarding state scales with the topology rather than the application-prefix count' — is contradicted by the paper's own data on edge PET growth. This is an internal inconsistency in the stated contribution, not merely a stylistic imprecision. The architecture's core-state benefit is real, so the paper need not be rejected; it should be accepted only if the claims are narrowed to core state or total per-router state is acknowledged to remain prefix-scaling. This is consistent with the reader's CONDITIONAL verdict, hence no adjustment is needed.","tokens_in":15526,"tokens_out":9602,"duration_ms":87241,"concrete_test":"Compute total network forwarding state = sum over all routers of (FIB + PET + BIFT) for AS 1755 at U=100 and U=1000 using the Section IV-B values: core routers (111) at 356.26 entries each, edge routers (61) at 645.92 and 1725.92 respectively, plus the reported control state. Total SRAN state increases by approximately 61 × (1725.92 − 645.92) = 65,880 entries as U grows from 100 to 1000. Since total state grows with prefix count, the claim that 'network forwarding state scales with topology rather than application-prefix count' is refuted as stated. The corrected claim should explicitly restrict the scaling statement to core routers.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and Contribution 3 claim that 'routers' forwarding states scale with the network topology, not the application prefixes.' This is false for the system as a whole. In SRAN, every edge router's PET contains an entry for every announced prefix (the PSD is replicated to all edge routers), so per-edge state is O(P). Fig. 6 (right) shows average edge state increasing from 645.92 entries at U=100 to 1725.92 at U=1000, a linear slope of 1.2 entries per unicast prefix. Total network state is roughly E*P (plus core constants), with E=61 in AS 1755; it still scales linearly with the application-prefix count, just with a smaller constant than the baseline's R*P (R=172). The core-only state is indeed topology-bound, but the paper's stated claim is about 'network forwarding state' without qualification. The evaluation's use of per-router state separated by core/edge does not support the unqualified claim; it supports 'core forwarding state scales with topology, while edge state is moved from all routers to edge routers.' This is more than a wording issue: the headline contribution is quantitatively different from what is claimed — a constant-factor reduction in the slope of state growth (61/172 ≈ 35% of baseline slope), not a change from prefix-scaling to topology-scaling.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"SRAN is an NDN routing architecture that applies the Map-and-Encap principle to separate name-prefix reachability from topological reachability. Edge routers maintain a Prefix Egress Table (PET) that maps name prefixes to egress routers; the PSD control-plane database is replicated across edge routers via NDN Sync and secured with NDN trust schemas. Core routers forward only on egress-router names (unicast) or BIER bitstrings (multicast), so core FIB/BIFT state is topology-bound. The paper reports a Go/NDNd prototype and evaluates it on Sprint PoP and Rocketfuel AS 1755, claiming reduced convergence overhead and core state independent of application-prefix count, while acknowledging that edge state still grows with prefixes.","tokens_in":15903,"tokens_out":7091,"duration_ms":62586,"significance":"The core architectural idea is well motivated: NDN's name-based forwarding indeed creates prefix-driven state in the entire network, and moving the prefix-to-locator mapping to the edge is a natural application of Map-and-Encap. The paper's concrete contributions include a unified unicast/multicast design using BIER, the use of NDN's native Sync and security to build the mapping dissemination plane, and an actual implementation with state-count measurements on real topologies. The paper also honestly reports that edge state grows with prefixes in Section IV.B, which is a useful data point. However, the abstract and Contribution 3 overstate the result as total state scaling with topology rather than prefix count, and the evaluation leaves key dynamic behaviors unexamined. With appropriate qualification and additional evaluation, the work would be a solid systems contribution; in its current form the headline claim is not supported by the paper's own data.","major_comments":[{"comment":"The abstract and Contribution 3 state that forwarding state scales with the network topology, not the application prefixes. This is contradicted by Fig. 6 (right), where SRAN edge PET state grows linearly from 645.92 to 1725.92 entries as the unicast prefix count U goes from 100 to 1000. Section IV.B itself says 'SRAN does not remove prefix state from the network; it moves that state to the edge.' Since the PSD is replicated to all edge routers, per-edge PET state is O(P) and total state is O(E·P) plus topology-bound core state. The supported claim is core-only topology-bounded state and a constant-factor reduction in total state slope (E vs. R), not prefix-independence of overall state. Please rewrite the headline claims to distinguish core, edge, and total state, and give the total-state formula.","section":"Abstract; §I Contribution 3; §IV.B 'The Effect of Application Dynamics' and Fig. 6"},{"comment":"Algorithm 1 drops a vanilla Interest when PET.Lookup returns no E∪F, so an Interest for a prefix that has not yet propagated to the ingress edge, or that has just been withdrawn or migrated, is dropped at ingress. The evaluation measures convergence latency and packet overhead but does not quantify the fraction of Interests dropped during convergence windows. Because the core-state benefit depends entirely on edge-side mapping availability, this miss path should be evaluated (e.g., by injecting Interests during announce/migrate events at different rates) and its behavior justified. Without such a study, the claim of 'real-time' adaptation is not substantiated.","section":"§III-B, Algorithm 1 lines 3–5; §IV.B"},{"comment":"For AS 1755, the text reports that both announcement and migration converge in exactly 133 ms for both SRAN and the baseline, and only packet overhead differs. This exact equality suggests the ns-3 control-plane model does not differentiate the two designs in ways that affect convergence (e.g., Sync cycles, retransmissions, FIB update propagation). The convergence evidence on the target topology is therefore non-informative. Either model the control-plane dynamics more faithfully or explicitly state that convergence is pinned to a fixed topology-diameter delay, and separate the convergence claim from the overhead result.","section":"§IV.B 'Prefix-Change Convergence and Overhead (AS 1755)'; Tab. IV"}],"minor_comments":[{"comment":"No artifact URL or repository is provided for the 'Golang-based SRAN implementation,' which limits reproducibility. The authors should add an availability statement with instructions for obtaining the code and running the experiments.","section":"§IV.A"},{"comment":"The BIFT table for BFR-id 5 lists both BFR-id 4 and BFR-id 6 with BFR-NBR R6 and the same F-BM, and no row for BFR-id 5 itself. Please clarify whether rows are per BFR-NBR or per BFR-id, and how the F-BM for R6 can contain both bits 4 and 6.","section":"Fig. 3b"},{"comment":"The BitString length and NDNLPv2 encapsulation overhead are not discussed. A 172-router domain requires at least 172 bits; larger Rocketfuel topologies (624, 960 routers) would need proportionally larger headers. One sentence on header-size implications would help.","section":"§III-C and §IV-A"},{"comment":"Algorithm 1 appears in §III-B for unicast, while §III-C refers to 'Alg. 2' for the complete process. Make the numbering and cross-reference consistent so readers know where the multicast-specific algorithm is defined.","section":"§III-B / §III-C"}],"recommendation":"major_revision","confidential_remarks":"This is a well-motivated systems paper with an honest internal evaluation and a useful state-placement result, but the abstract and Contribution 3 overclaim the scalability property in a way that the paper's own Fig. 6 contradicts. The miss-path behavior during convergence is an unexamined load-bearing point, and the ns-3 convergence result (identical 133 ms for both designs) undermines the real-time claim. These issues are fixable within the scope of the manuscript, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the take: SRAN is a real architecture with real code and a real evaluation, and it deserves a serious referee. But the headline claim in the abstract — that “network forwarding state scales with the topology rather than the application-prefix count” — is only true for the core. Total network state still grows linearly with prefixes: every edge router's PET holds an entry for every announced prefix, so the scaling law is O(E·P), not topology-bound. The stress-test note is right about this. Fig. 6's right panel shows edge state climbing from about 646 to 1726 entries as unicast prefixes go from 100 to 1000. What SRAN actually does is move prefix state from all routers to edge routers, cutting the slope from R·P to E·P (61 vs 172 on AS 1755) and making core state fixed. That's a meaningful constant-factor and architectural improvement, but it's not the change of scaling the abstract implies.\n\nWhat's genuinely new: SRAN is the first NDN routing design to combine L2.5 Interest encapsulation to egress routers with BIER bitstrings for multicast and native NDN Sync for disseminating prefix-to-egress mappings. The PET/PSD split is clean, and using NDN's existing security and Sync is elegant — the control plane falls out of the substrate rather than being bolted on. The implementation in NDNd is substantial, and the evaluation covers an all-edge topology and the intended core/edge case, with state snapshots, scaling sweeps, and analytical extrapolation to larger Rocketfuel topologies. The paper is also honest in the body: Section IV explicitly says SRAN “does not remove prefix state from the network; it moves that state to the edge.” The problem is the abstract and contributions don't carry that nuance.\n\nSoft spots, in proportion: first, the scaling claim needs rewording in the abstract and Contribution 3; the data supports “core state is topology-bound” and “edge state absorbs prefix growth with a smaller slope.” Second, the AS 1755 convergence reporting is coarse: both announcement and migration come out at exactly 133 ms, and no baseline convergence is shown there — either the ns-3 model isn't distinguishing the control planes or the comparison got lost. Third, no code or artifact is released, so independent reproduction is impossible. Fourth, the miss path is unevaluated: Alg. 1 drops an Interest if the PET lacks a mapping, and the paper never quantifies traffic dropped during PSD convergence or after a withdrawal. Fifth, the security analysis is qualitative, which is acceptable at this stage, though “no additional protocol” overstates things since the PSD is a new control-plane component built on Sync.\n\nBottom line: this paper is aimed at NDN/ICN researchers, and they should read it. It's a solid candidate for peer review — a revision that fixes the claim, releases the code, and reports baseline convergence on AS 1755 would be strong. I'd send it to referees.","headline":"A genuine NDN routing architecture with substantial implementation and evaluation, but the flagship claim that forwarding state scales with topology is only true for core routers; total state still grows with prefixes, just at the edge.","tokens_in":16370,"tokens_out":4741,"would_cite":true,"duration_ms":37406,"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":"SRAN separates name-prefix reachability from topological reachability, moving prefix mappings to edge routers so NDN core forwarding state scales with topology, not application names.","keywords":["Named Data Networking","Map-and-Encap","routing scalability","BIER","multicast forwarding","prefix egress table","NDN routing","control plane dissemination"],"falsifier":"Run the AS 1755 forwarding-state measurement with 10,000 announced application prefixes and inspect the FIB at a core router: if any application-prefix entry appears, SRAN's claim that core state is topology-bound fails. The paper's forwarding-state sweep stops at 1,000 prefixes, so this is a direct test of the flat-core prediction.","tokens_in":15439,"feed_emoji":"🌐","tokens_out":5876,"duration_ms":43706,"temperature":0.7,"pith_summary":"SRAN's central claim is that Named Data Networking can escape the routing-scalability trap that comes from fetching data by name. The paper proposes separating name-prefix reachability from topological reachability: edge routers maintain a Prefix Egress Table that maps application prefixes to egress routers, and the core forwards only on router-level names or BIER bitstrings. If true, core forwarding state stays fixed by topology — on the AS 1755 topology, 172 FIB and 172 BIFT entries while application prefixes grow to 1,200 — while prefix changes converge within roughly 99–133 ms with less control traffic than a broadcast baseline. A sympathetic reader would care because this is a concrete path to a data-centric network whose core does not carry per-application state, and the same mapping table serves unicast and multicast.","feed_headline":"Core NDN routers stay flat as app prefixes grow","feed_subtitle":"SRAN maps names to edge routers, so core FIB stays at 172 entries even at 1,200 prefixes.","key_machinery":"The load-bearing mechanism is the Prefix Egress Table (PET), a prefix-to-router mapping that lives only at edge routers, together with the Prefix State Database (PSD), the synchronized control-plane store that populates PETs. Ingress edge routers do a two-phase lookup — PET first to choose an egress router, then FIB on that router's name — and encapsulate the original Interest in an NDNLPv2 header carrying either the egress-router identifier or a BIER BitString. Core routers forward on that header using a topology-bound FIB or BIFT, so the per-hop table stays independent of the number of application prefixes.","core_discovery":"The paper's discovery is that NDN's routing state can be made to scale with the physical topology rather than the application namespace by applying Map-and-Encap. In SRAN, application prefixes are announced to edge routers, stored in a synchronized Prefix State Database, and materialized into a Prefix Egress Table at the edge; an ingress router encapsulates each Interest with either the chosen egress router's name (unicast) or a BIER BitString marking all target egress routers (multicast). Core routers therefore never see application names and consult only topology-bound FIB/BIFT tables. Evaluation on the Sprint PoP and AS 1755 topologies shows the expected split: core FIBs stay at 172 entri","pith_inferences":["The design's benefit assumes the edge PET already contains a mapping for every Interest's name; during the PSD-synchronization window after a new announcement, an Interest for a not-yet-mapped prefix is dropped (Algorithm 1, lines 3–5). A testable extension would quantify the fraction of early traffic lost and whether application retries absorb it.","SRAN moves prefix state to the edge rather than eliminating it: edge PETs grow linearly with announced prefixes, so the trade-off is acceptable only when the edge is willing to absorb that memory. For networks with many edge routers, the aggregate edge state is the real scaling limit.","The paper's recursion to inter-domain routing implies an AS-level PET/PSD, but the hard open problem there is trusting prefix-to-AS mappings across administrative domains — the paper sketches the nested design but does not evaluate its convergence or security.","Because multicast and unicast share the PET, one could extend SRAN to dynamic group membership by treating joins and leaves as prefix updates; the paper does not explore how fast such churn could propagate through PSD."],"forward_implications":["If SRAN is correct, core routers in an SRAN domain keep a fixed number of forwarding entries regardless of how many application prefixes are announced; on AS 1755 the FIB and BIFT both remain at 172 entries while application prefixes grow to 1,200, versus a baseline that grows linearly.","Prefix announcements and migrations converge close to ordinary NDN routing — medians of 99 ms and 125 ms on Sprint PoP and 133 ms on AS 1755 — so the scalability gain does not require sacrificing convergence speed.","BIER-based dissemination of prefix state cuts control-plane packet overhead compared with broadcast Sync, with median announcement overhead on AS 1755 dropping from 8.65 packets to 2.45 packets per link.","Multicast forwarding state is also topology-bound: the same BIFT serves all multicast prefixes, because the PET supplies the target egress set as a BitString.","The same map-and-encap structure can be applied recursively between ASes, with an AS-level PET mapping prefixes to egress ASes and transit ASes forwarding on AS-level reachability."],"fun_headline_variants":["SRAN keeps NDN core state tied to topology, not app names","NDN core scales with topology when app names are mapped to edge","Map-and-Encap: NDN routing state shrinks to topology size","SRAN: app prefixes stay at edge, core FIB stays flat","NDN core routers no longer track every app prefix"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"An ingress edge router must already have a PET entry for a name prefix when the first Interest for it arrives; if PSD synchronization has not yet delivered that mapping, the Interest is dropped rather than forwarded.","fun_headline_variants_meta":{"raw":{"variants":["SRAN keeps NDN core state tied to topology, not app names","NDN core scales with topology when app names are mapped to edge","Map-and-Encap: NDN routing state shrinks to topology size","SRAN: app prefixes stay at edge, core FIB stays flat","NDN core routers no longer track every app prefix"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000598,"raw_usage":{"total_tokens":2632,"prompt_tokens":739,"completion_tokens":1893,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":483,"completion_tokens_details":{"reasoning_tokens":1801}},"tokens_in":483,"tokens_out":1893,"duration_ms":16718,"temperature":1.0,"reasoning_tokens":1801,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T10:01:15.483772+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the AS 1755 forwarding-state measurement with 10,000 announced application prefixes and inspect the FIB at a core router: if any application-prefix entry appears, SRAN's claim that core state is topology-bound fails. The paper's forwarding-state sweep stops at 1,000 prefixes, so this is a direct test of the flat-core prediction.","supporting_citations":[],"review_version":1}