REVIEW 3 major objections 5 minor 20 references
Mathematical Modeling for Network Upgrades in Internet Service Provider Infrastructure
T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash
Pith's one-line read A queueing model predicts an ISP link's breaking point at about 30 users, and how upgrading past it removes the collapse.
desk verdict The paper's central upgrade-threshold claim is contradicted by its own equations: with the stated constants, rho hits 1 at N≈20, so the reported 30-user threshold and 2-second delays cannot come from the presented model. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the classic M/M/1 single-server queueing relation — one server, memoryless arrivals and service times — expressed in the paper as traffic intensity $\rho = S\lambda_{\text{total}}/B$ with queueing delay $\rho/(1-\rho)$ for $\rho<1$. This single formula makes delay grow without bound as the arrival rate approaches bandwidth, so the sharp delay spike at roughly 30 users is simply the point where $\rho$ approaches 1. The paper wraps this core in supporting equations for transmission delay $S/B$, propagation delay $L/v$, processing delay, server utilization, throughput, and packet drops, but the upgrade threshold itself comes from the $\rho/(1-\rho)$ blow-up.
What would settle it
Run a controlled test on a real 100 Mbps link with up to 50 users streaming Full HD video and record average delay as users are added one at a time. If the sharp delay escalation does not occur near 30 users, or if delay stays flat because bursty traffic multiplexes, the model's threshold claim fails. Directly measuring per-user request rates would also test the 417 requests/second assumption.
Extended reading notes
Core claim
The paper's central claim is that the network's behavior is governed by traffic intensity $\rho = S\lambda_{\text{total}}/B$, where $S$ is packet size, $\lambda_{\text{total}}$ is the arrival rate from $N$ users, and $B$ is bandwidth, with queueing delay given by $\rho/(1-\rho)$ when $\rho<1$ and infinite otherwise. Simulating these equations for the assumed workload — each user sends 417 requests per second, a 5 Mbps Full HD stream — shows that as $N$ approaches 30 on a 100 Mbps link, the queue saturates, delay escalates sharply to 2 seconds per user, queue drops reach the 1000-packet limit, and throughput collapses. In the upgraded network, increasing bandwidth to 1 Gbps and scaling the queue removes packet drops and keeps delay flat at about 0.02 seconds while throughput climbs linearly past 400 Mbps at 50 users. The paper interprets these results as evidence that the model can identify upgrade thresholds and guide ISP investment decisions.
Load-bearing premise
The prediction rests on the assumption that every user steadily sends 417 requests per second (a fixed 5 Mbps Full HD stream), so the queue sees a constant load; if real subscriber traffic is bursty or time-varying, the 30-user boundary may not hold.
Editorial extensions
If this is right
- An ISP could plug its own link speed, packet size, per-user demand, and server capacity into these equations to get a numeric user limit before buying equipment.
- Raising the link from 100 Mbps to 1 Gbps and scaling the queue is predicted to eliminate packet drops and keep delay near 0.02 seconds for at least 50 users.
- Server utilization reaches 100% at around 15 users in both configurations, so server capacity, not link bandwidth, becomes the first bottleneck after an upgrade.
- A quality target such as delay under 100 ms for real-time video can be inserted into the model to produce a concrete, application-specific upgrade trigger.
Reading between the lines
- Feeding real, bursty traffic traces into the same equations would likely shift or soften the 30-user cliff, because aggregation of many streams smooths the load the queue sees.
- The model could be inverted to solve for the bandwidth needed to hold delay under a given target at a target subscriber count, turning a threshold detector into a capacity planner.
- The sharp threshold is tied to the constant 5 Mbps per-user assumption; with variable-rate video codecs, the saturation point would vary with content, so the 30-user number is illustrative rather than universal.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes an analytical, M/M/1-style queueing model for ISP network upgrades. It defines traffic intensity as rho = S*lambda_total/B, queuing delay as rho/(1-rho), and several related delay, utilization, throughput, and packet-drop formulas. Using a baseline bandwidth of 100 Mbps, a packet size of 1500 bytes, and a per-user request rate of 417 requests/second, the paper claims that simulations identify an upgrade threshold around 30 users, after which delay escalates sharply and throughput collapses, and that upgrading to 1 Gbps eliminates these bottlenecks. The paper frames its contribution as a mathematical tool that ISPs can use to predict when infrastructure upgrades are needed.
Significance. The topic is practically relevant, and the paper's explicit equations and parameter choices at least make the modeling assumptions checkable. If the reported threshold were a genuine, validated prediction, the approach could be useful for capacity planning. However, the paper provides no empirical validation, no fitted parameters, no comparison against measured ISP traffic, and no reproducibility artifacts such as code or data. More importantly, the reported numerical results are internally inconsistent with the model's own equations: the stated constants make rho reach 1 at about 20 users, not 30, and the reported server utilization reaches 100% at 15 users despite the stated server capacity of 50 users. These load-bearing inconsistencies mean the central claim of identifying upgrade thresholds is not supported by the presented model.
major comments (3)
- [Section III-D, Eqs. (1)-(3) and Section IV] With B = 100 Mbps, S = 12000 bits, and lambda_user = 417 requests/second, Eq. (2) gives lambda_total = 417N and Eq. (1) gives rho = 0.05004N. Equation (3) then yields infinite queuing delay for N >= 20. The text of Section IV states that delay remains negligible below 30 users, rises sharply only after 30, and reaches a finite peak of 2 seconds. This is not the output of the presented equations. If the simulation instead used a finite queue of 1000 packets, then the model evaluated is M/M/1/K, not Eq. (3), and the paper does not provide the M/M/1/K delay formula or any derivation of the claimed threshold. The headline numerical result is therefore unsupported by the model as stated.
- [Section IV, Figures 5 and 9; Section III-C] The paper reports that server utilization reaches 100% at around 15 users in both the original and upgraded networks, but Section III-C states that the maximum server capacity is 50 users. Equation (8) defines server utilization abstractly as a ratio of arrival-related and service-related quantities but does not connect it to the stated 50-user capacity or to N. If utilization is N/50, it reaches 100% only at N = 50; if utilization equals rho from Eq. (1), it reaches 100% near N = 20. The reported curve at N = 15 is inconsistent with both interpretations, which makes the 'server capacity' parameter non-functional and undermines the interpretation of the upgrade's effect.
- [Section III-D and Section IV] The claimed upgrade threshold is not predicted by the model but is forced by the choice of constants. Since rho = 0.05004N, any threshold in N corresponds directly to the chosen B, S, and lambda_user, and the upgraded scenario is simply the arithmetic of reducing rho by a factor of 10. The paper provides no measured traffic traces, no external benchmark, no sensitivity analysis, and no goodness-of-fit statistics, so there is no evidence that the threshold is meaningful for real ISP networks. This circularity, together with the internal contradiction noted above, means the central claim of predictive upgrade timing is not established.
minor comments (5)
- [Title page] The manuscript retains the IEEE template placeholder 'REPLACETHISLINEWITHYOURPAPERIDENTIFICATIONNUMBER' and the boilerplate first-footnote text, indicating that the submission was not prepared for review.
- [Section IV] The in-text figure references are inconsistent with the figure captions: the text says 'Throughput, as illustrated in Figure 8' and 'Queue drops ... as evidenced in Figure 8', but Figure 8 is labeled 'Queue Drops' and the preceding figure, Figure 7, is labeled 'Throughput'. The references should be corrected to match the figure numbering.
- [Section III-D] Equations (10)-(12) are typeset with broken notation, such as '??? = ??? (0, min (...))', and Eq. (8) uses undefined symbols for arrival rate and service rate. All variables should be defined consistently with Eqs. (1)-(7).
- [Section II] The literature review cites references [17]-[20], which concern energy harvesting, smart metering, and industrial IoT security; these do not appear relevant to network upgrade modeling and should be replaced with appropriate prior work or removed.
- [Abstract] The abstract states that 'simulations prove the model's usefulness', but the paper provides no confidence intervals, goodness-of-fit measures, or comparison against alternative models; this claim should be softened or supported.
Circularity Check
The paper's headline upgrade threshold and its claimed proof of model usefulness reduce to its own assumed equations; no independent data or benchmark is introduced.
-
other
[Abstract and Section III.E (Simulation Program)]
"This paper’s simulations prove the model’s usefulness in pinpointing such thresholds so that ISPs can prepare their upgrades and prevent performance constrictions on schedule. ... The network equations in this study were simulated using Python, chosen for its flexibility and efficiency in scientific computing. The program integrates the mathematical formulas for traffic modeling, queuing dynamics, delays, throughput, and utilization."
The claimed proof is the Python program that implements the same Eqs. (1)-(11) being evaluated. No external traffic data, independent measurements, or benchmark results are introduced. Consequently, the simulation's output is the model's own equations evaluated back at the reader; saying it 'proves the model's usefulness' is equivalent to saying the equations produce their own outputs. This makes the central validation circular.
-
self definitional
[Section III.C-D (Assumptions and Eqs. 1-3) and Section IV (Fig. 2)]
"Traffic Intensity: ? = S.λ_total / B (1) ... Queuing Delay: ... ?_queue = ρ/(1−ρ), ρ<1; ∞, ρ≥1 (3) ... In the original configuration (Figures 2–5), the total delay per user remained negligible for user counts below 30 ... beyond this threshold, the delay escalated sharply, reaching a peak of 2 seconds per user."
Using the paper's constants (S=12000 bits, B=100 Mbps, per-user rate 417 req/s) in Eq. (1), ρ = N·12000·417/100,000,000 = 0.05004N; Eq. (3) then gives infinite queuing delay at N≈20. The 'threshold around 30 users' is therefore not an empirical or independently derived finding: it is a direct arithmetic consequence of the chosen constants, and it is actually inconsistent with the quoted equations. The upgraded-network improvement is likewise just the arithmetic of increasing B from 100 Mbps to 1 Gbps. The predictions carry no information beyond the assumptions, and the paper's simulation cannot validate them.
full rationale
The self-citations in [17]-[20] are not load-bearing: they support general framing rather than the mathematical results. However, the central 'predictions' are circular in a different sense: the delay and throughput curves are generated by the same equations that define the model, and the threshold is an arithmetic consequence of the assumed S, B, and per-user rate. The paper's own Eq. (3) places saturation at about 20 users for its constants, not at the reported 30, which is an internal inconsistency; that is a correctness defect, but it reinforces that the headline number is not an independent result. The absence of external measurements or benchmarks means the simulations cannot validate the model's usefulness, so the claimed proof is self-referential.
Assumptions & free parameters
free parameters (7)
- Per-user request rate lambda_user =
417 requests/second
- Bandwidth B (baseline) =
100 Mbps
- Server capacity =
50 users
- Queue limit =
1000 packets
- Upgraded bandwidth =
1 Gbps
- Queue scaling factor =
5
- Propagation cable length L =
90 m
assumptions (3)
- domain assumption Arrivals follow a Poisson process and service times are exponential (M/M/1 queue), so Eq (3) for queuing delay applies.
- domain assumption The network can be modeled as a single bottleneck queue where all traffic aggregates to the bandwidth B.
- ad hoc to paper The server's processing rate is implicitly tied to 'server capacity = 50 users' with no explicit service rate mu defined.
Cite this review
Pith. "Pith review of Mathematical Modeling for Network Upgrades in Internet Service Provider Infrastructure." pith.science (2026). https://pith.science/paper/X45IZFYN
@misc{pith2026250200787,
author = {Pith},
title = {Pith review of: Mathematical Modeling for Network Upgrades in Internet Service Provider Infrastructure},
year = {2026},
howpublished = {\url{https://pith.science/paper/X45IZFYN}},
note = {Machine review of arXiv:2502.00787}
}
read the original abstract
The ongoing growth of the need for superior Internet services creates great pressure on the ISPs as to the accurate estimation of network upgrade need.
Reference graph
Works this paper leans on
-
[1]
Performance modeling and analysis of network firewalls,
K. Salah, K. Elbadawi, and R. Boutaba, “Performance modeling and analysis of network firewalls,” IEEE TransactionsonNetworkand ServiceManagement,vol.9,no. 1, pp. 12–21, Mar. 2012, doi: 10.1109/TNSM.2011.122011.110151
arXiv 2012
-
[2]
F. Carpio, W. Bziuk, and A. Jukan, “Scaling Migrations and Replications of Virtual Network Functions based on Network Traffic Forecasting,” May 2021, doi: 10.1016/j.comnet.2021.108582
-
[3]
An Adaptive Framework for Generalizing Network Traffic Prediction towards Uncertain Environments
A. Downey, E. Tuna, and A. Soysal, “An Adaptive Framework for Generalizing Network Traffic Prediction towards Uncertain Environments,” Nov. 2023, [Online]. Available:http://arxiv.org/abs/2311.18824
work page Pith review arXiv 2023
-
[4]
Digital Twin for Networking: A Data-driven Performance Modeling Perspective
L. Hui, M. Wang, L. Zhang, L. Lu, and Y. Cui, “Digital Twin for Networking: A Data-driven Performance Modeling Perspective,” Jun. 2022, [Online]. Available: http://arxiv.org/abs/2206.00310
work page Pith review arXiv 2022
-
[5]
Deeper and Wider Networks for Performance Metrics Prediction in Communication Networks
A. Liu, S. Liu, and X. Pei, “Deeper and Wider Networks for Performance Metrics Prediction in Communication Networks,” Dec. 2023, [Online]. Available: http://arxiv.org/abs/2401.00429
work page Pith review arXiv 2023
-
[6]
Rethinking Data-driven Networking with Foundation Models: Challenges andOpportunities,
F. Le, M. Srivatsa, R. Ganti, and V. Sekar, “Rethinking Data-driven Networking with Foundation Models: Challenges andOpportunities,”inHotNets2022-Proceedingsofthe2022 21stACM Workshop on HotTopicsin Networks,Association for Computing Machinery, Inc, Nov. 2022, pp. 188–197. doi: 10.1145/3563766.3564109
arXiv 2022
-
[7]
RouteNet-Erlang: A Graph Neural Network for Network Performance Evaluation,
M. Ferriol-Galmes et al., “RouteNet-Erlang: A Graph Neural Network for Network Performance Evaluation,” Proceedings - IEEE INFOCOM, vol. 2022-May, pp. 2018–2027, 2022, doi: 10.1109/INFOCOM48880.2022.9796944
arXiv 2022
-
[8]
Learning-Based Network Performance Estimators: The Next Frontier for Network Simulation,
K. Shen and B. Li, “Learning-Based Network Performance Estimators: The Next Frontier for Network Simulation,”IEEENetw,vol.37,no.4,pp.97–103,Jul.2023, doi:10.1109/MNET.013.2300053
Show all 20 references
-
[9]
Digital Twin for Networking: A Data-Driven Performance Modeling Perspective,
L. Hui, M. Wang, L. Zhang, L. Lu, and Y. Cui, “Digital Twin for Networking: A Data-Driven Performance Modeling Perspective,” IEEE Netw, vol. 37, no. 3, pp. 202–209, May 2023,doi:10.1109/MNET.119.2200080
2023 doi
-
[10]
Data-driven Predictive Latency for 5G: A Theoretical and Experimental Analysis Using Network Measurements,
M.Skocajetal.,“Data-driven Predictive Latency for 5G: A Theoretical and Experimental Analysis Using Network Measurements,” Jul. 2023, doi: 10.1109/PIMRC56721.2023.10293861
2023
-
[11]
Performance prediction and enhancement of 5G networks based on linear regression machine learning,
M. Malekzadeh, “Performance prediction and enhancement of 5G networks based on linear regression machine learning,” EURASIP J Wirel Commun Netw, vol. 2023,no.1,Dec.2023,doi:10.1186/s13638-023-02282-z
2023 doi
-
[12]
Network Traffic Prediction Model in a Data-Driven Digital Twin Network Architecture,
H. Shin, S. Oh, A. Isah, I. Aliyu, J. Park, and J. Kim, “Network Traffic Prediction Model in a Data-Driven Digital Twin Network Architecture,” Electronics (Switzerland), vol. 12,no.18,Sep.2023,doi:10.3390/electronics12183957
2023 doi
-
[13]
Formal MethodsforNetworkPerformanceAnalysis,
M. T. Arashloo, R. Beckett, and R. Agarwal, “Formal MethodsforNetworkPerformanceAnalysis,”in20thUSENIX Symposium on Networked Systems Design and Implementation(NSDI23),Boston,MA:USENIXAssociation, Apr. 2023, pp. 645–661. [Online]. Available: https://www.usenix.org/conference/n...
2023
-
[14]
Towards a Unified Method for Network DynamicviaAdversarialWeightedLinkPrediction,
M. Qin, “Towards a Unified Method for Network DynamicviaAdversarialWeightedLinkPrediction,”Jan.2024, [Online].Available:http://arxiv.org/abs/2401.03444
2024 arXiv
-
[15]
AnEfficient Data-DrivenTrafficPredictionFrameworkforNetworkDigital Twin,
H.Nan,R.Li,X.Zhu,J.Ma,andD.Niyato,“AnEfficient Data-DrivenTrafficPredictionFrameworkforNetworkDigital Twin,” IEEE Netw, vol. 38, no. 1, pp. 22–29, Jan. 2024, doi: 10.1109/MNET.2023.3335952
2024
-
[16]
Modeling configuration-performance relation in a mobile network: a data-driven approach,
M. Panek, I. Jabłoński, and M. Woźniak, “Modeling configuration-performance relation in a mobile network: a data-driven approach,” Jul. 2024, [Online]. Available: http://arxiv.org/abs/2407.06702
2024 arXiv
-
[17]
E-proctoringsystems:A review on designing techniques, features and abilities against threatsand attacks,
H.M.MohammedandQ.I.Ali,“E-proctoringsystems:A review on designing techniques, features and abilities against threatsand attacks,”Quantum JournalofEngineering,Science andTechnology,vol.3,no.2,pp.14–30,2022
2022
-
[18]
Design, implementation & optimization of an energy harvesting system for VANETS’road side units (RSU)’,
Q. I. Ali, “Design, implementation & optimization of an energy harvesting system for VANETS’road side units (RSU)’,” IET Intell. Transp. Syst, vol. 8, no. 3, pp. 298–307, 2014
2014
-
[19]
An embedded and intelligentanomalypowerconsumptiondetectionsystembased onsmartmetering,
S. Lazim Qaddoori and Q. I. Ali, “An embedded and intelligentanomalypowerconsumptiondetectionsystembased onsmartmetering,”IETWirelessSensorSystems,vol.13,no. 2,pp.75–90,2023
2023
-
[20]
Anefficientsecuritymodel forindustrialinternetofthings(IIoT)systembasedonmachine learning principles,
S.L.QaddooriandQ.I.Ali,“Anefficientsecuritymodel forindustrialinternetofthings(IIoT)systembasedonmachine learning principles,”Al-Rafidain Engineering Journal (AREJ), vol.28,no.1,pp.329–340,2023
2023
Reviewed August 9, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.