Bus-Mounted Edge Servers for Urban IoT

• Bus-mounted edge servers are mobile computing platforms on transit buses that extend dynamic computation and connectivity to urban IoT applications.
• They leverage predictable bus mobility to facilitate smart city data collection, real-time analytics, and vehicular network support, reducing reliance on static infrastructures.
• Advanced placement algorithms and dynamic server selection optimize urban coverage and minimize network delays while addressing infrastructural and resource constraints.

Bus-mounted edge servers are mobile computing platforms installed on transit buses, designed to extend edge computing resources throughout urban environments. By leveraging the predictable mobility of public buses, these servers can provide dynamic computation, storage, and connectivity services to support smart city Internet-of-Things (IoT) deployments and vehicular network applications. This paradigm addresses both infrastructural limitations and the spatial-temporal dynamics of urban demand, offering an alternative or complement to static edge infrastructure such as Road Side Units (RSUs) or fixed base stations.

1. System Architectures and Operational Modes

Bus-mounted edge servers integrate hardware, connectivity, and software platforms within transit buses to provide localized computing and data relay capabilities. Three distinct architectural modes have been examined:

a. Smart City Data Mule Infrastructure:

• Small, delay-tolerant IoT sensors are co-located with bus stops and buffer data until an opportunistic contact with a passing bus.
• Each bus carries a wireless radio and periodically uploads collected sensor data when in proximity to fixed bus-stop gateways equipped with edge servers, which maintain always-on backhaul to the cloud or perform preliminary preprocessing (Madamori et al., 2021).

b. Mobile Edge Computation for IoV:

• Buses are equipped with edge servers possessing defined computation capacity ($C^{bus}$, tasks/slot) and a coverage radius $R$ (km).
• As buses follow known trajectories, they dynamically serve vehicular and urban device computation/communication requests as they traverse city grids (Li et al., 5 Dec 2025).

c. Real-time Onboard Analytics:

• Edge servers process high-frequency telemetry (e.g., GPS, operational data) directly on the bus, offloading only relevant or aggregated insights to the cloud via cellular backhaul, thereby reducing network load (Cao et al., 2017).

Hardware Example: The Cisco IR829 router, industrial-grade and environmental-resistant (–40 ℃ to +60 ℃), hosts an Intel Atom C2308 CPU, 2 GB RAM, and dual-SIM 4G LTE networking, supporting both Linux-based analytics and local routing (Cao et al., 2017).

2. Mathematical Models for Placement and Coverage

a. Static Gateway (Bus Stop) Selection

The minimal coverage-budget problem, termed Maximal Sensor Coverage (MSC), is modeled as a set-cover problem: activate the minimal set $G$ of bus stops as gateways such that every bus route is covered. Given $R$ (bus routes), $S_\ell$ (routes at stop $\ell$), binary variables $x_\ell$ indicate gateway activation:

$$\min_{x} \sum_{\ell} x_\ell \quad \text{s.t.}\;\; \sum_{\ell:\,r\in S_\ell} x_\ell \ge 1 \ \forall\,r\in R,\ \ x_\ell\in\{0,1\}$$

(Madamori et al., 2021).

b. Bus-Mounted Server Selection (Dynamic Coverage)

The optimization seeks to maximize the total covered demand across all time slots $T$ and grids $\mathcal{G}$. Decision variables $s_j$ (bus $j$ equipped: $0/1$), $\theta_{i,j}(t)$ (demand point $i$ at $t$ served by bus $j$):

$$\max \mathrm{CDP} = \sum_{t=1}^T \sum_{j=1}^M s_j \sum_{i=1}^{N(t)} \theta_{i,j}(t)$$

Subject to: \begin{align*} &\sum_{j=1}^M s_j = K \ &\sum_{i} \theta_{i,j}(t) \le C^{bus}\cdot s_j \ \forall j, t \ &\theta_{i,j}(t)=0 \text{ if }|pos(l_i^{dp}(t))-pos(l_j^{bus}(t))|>R \ &s_j,\theta_{i,j}(t)\in{0,1} \end{align*} This is an NP-hard integer quadratic program (Li et al., 5 Dec 2025).

c. Delay Metric in Opportunistic Collection

For sensor $s$ at stop $v$ and a gateway set $G$: $$D_s(G)= \begin{cases} \min_{u\in G}\bigl[t_{\mathrm{wait}}(v)+\tau(u,v)\bigr] & \text{if any } u\in G \text{ reachable}\ T & \text{otherwise} \end{cases}$$ with average network delay $\bar D(G)=\frac1{|S|}\sum_{s\in S} D_s(G)$ (Madamori et al., 2021).

3. Placement Algorithms and Computational Approaches

a. Greedy Set-Cover Algorithms:

• MSC-Greedy iteratively selects the stop covering the most yet-uncovered routes, achieving logarithmic approximation (Madamori et al., 2021).

b. Submodular Optimization for Delay:

• GREEDY-MDD and CELF-MDD heuristics incrementally add stops with maximal marginal reduction in $\bar D(G)$. CELF-MDD reduces influence evaluation calls by up to 75% versus naive greedy, yielding order-of-magnitude speedup on large transit graphs (Madamori et al., 2021).

c. Greedy Bus Selection for Dynamic Coverage:

• For the dynamic bus selection, the greedy heuristic selects at each iteration the bus that, given $C^{bus}$ and $R$, maximizes coverage of yet-unserved demand points, recursing until the bus budget $K$ is exhausted:

Input: Bus trajectories ℛ, demand points {𝒫(t)}, radius R, capacity C^{bus}, K Output: S ← ∅ for iter in 1…K: for each remaining bus j in ℛ: for t = 1…T: P_j(t) = up to C^{bus} closest points in 𝒫(t) within radius R C_j = sum_t len(P_j(t)) x = argmax_j C_j S = S ∪ {x} Remove x from ℛ; remove all P_x(t) from 𝒫(t) ∀t if all 𝒫(t) empty, break return S

Complexity is $O(K \cdot M \cdot T \cdot \log N)$ (Li et al., 5 Dec 2025).

4. Evaluation, Quantitative Results, and Comparative Analysis

a. Static Gateways on Public Transit Networks

For two US cities:

• Chapel Hill Transit (CHT, 26 routes): MSC-Greedy required only 4 gateways, while in-degree and betweenness centrality required 18 and 25 respectively (over 77% reduction).
• Louisville (TARC, 46 routes): 13 (MSC) vs. 237 (IC) and 459 (BC).
• CELF-MDD yielded up to 20–45 min lower delivery delay than centrality-based baselines for $k=5$ (CHT) and $k=9$ (TARC), after which additional gateways showed diminishing returns (Madamori et al., 2021).

b. Bus-Mounted Server Feasibility in Shanghai

• At $R=0.8$ km, $C^{bus}=50$, selecting $K=1,000$ (of 2,100) buses: area coverage = 50%, point coverage = 53%, point coverage with cap = 40%.
• At $K=500$, greedy approach yields 35.7% coverage vs. 23.3% (random), an improvement of 53%.
• Higher $C^{bus}$ increases point coverage (30: 40%; 50: 51%; 100: 62%).
• Combined with 30% of base stations, 500 bus-mounted edge servers eliminated ≥90% of urban demand hotspots during peak loads (Li et al., 5 Dec 2025).

c. Onboard Analytics and Real-time Descriptive Functions

• Sampled bus telemetry at 5 s intervals, totaling 17,280 tuples per day.
• Descriptive analytics (stops, moves, trip durations) executed in-situ; only trip and day-level summaries sent to the cloud, reducing backhaul costs.
• Example: on Feb 14, minimum trip duration 897 s, maximum 13,468 s; Feb 14–15 averaged 47–61 trips, 214 moves/trip, 2,693–3,056 stops/trip (Cao et al., 2017).

5. System Limitations, Trade-offs, and Best Practices

Limitations

• Static placement and server selection do not adapt to intra-day fluctuations in demand or traffic (Li et al., 5 Dec 2025).
• Energy and maintenance costs, server on/off scheduling, and interference are not modeled.
• Edge servers’ resource constraints (RAM, CPU, storage) require analytics partitioning and demand explicit queue management (Cao et al., 2017).
• Real-time communication is not guaranteed for all application classes; models target delay-tolerant or batch data transfer (Madamori et al., 2021).

Trade-offs

• Server budget ($K$ and $C^{bus}$) exhibit diminishing returns: increasing $K$ from 500 to 1,000 yields 15–20% coverage improvement, but further increases have less marginal effect (Li et al., 5 Dec 2025).
• Pareto-optimal combinations ($K$, $C^{bus}$) allow operators to tune coverage and quality-of-service (QoS) (Li et al., 5 Dec 2025).
• Lightweight, period-driven analytics lower resource cost and prioritize actionable, delay-tolerant results (Cao et al., 2017).

Best Practices

• Employ robust pre-processing for bus telemetry streams to counteract missing, duplicate, or invalid data.
• Partition computation to maximize local result utility while minimizing raw data transmission (Cao et al., 2017).
• Monitor wireless connectivity continuously and include adaptive re-try logic in dissemination modules.

6. Future Directions and Open Challenges

Proposed future work includes:

• Multi-objective placement algorithms encompassing energy usage, latency, and operational cost (Li et al., 5 Dec 2025).
• Online and adaptive bus/server selection to match temporal demand patterns in real time.
• Integrated orchestration of bus-mounted servers with static RSUs and macro-cellular networks for end-to-end resource optimization.
• Enhanced models factoring in bus schedules, vehicle-to-vehicle (V2V) link quality, and battery/power constraints.
• Empirical evaluation of interference, handoff performance, and the interaction between mobile and static edge infrastructure.

This suggests that the domain is moving towards holistic, adaptive, and data-driven orchestration of edge resources in urban vehicular environments, with bus-mounted nodes providing unique, cost-effective, and flexible computation compared to solely static deployments.

---

References:

• [Enabling Opportunistic Low-cost Smart Cities By Using Tactical Edge Node Placement, (Madamori et al., 2021)]
• [Are Bus-Mounted Edge Servers Feasible?, (Li et al., 5 Dec 2025)]
• [Developing an edge computing platform for real-time descriptive analytics, (Cao et al., 2017)]