Computer Engineering is coming soon

We're still writing and reviewing the notes for every subject, and we'll open it when every subject is ready. Meanwhile you can practise aptitude, ask Aria about any subject, or see the branches that are ready.

Edge computing vs distributed cloud

Compare edge computing (compute near data sources) with distributed cloud (cloud provider spreads cloud services geographically), focusing on latency, placement, management and use-cases for engineers.

Drafted with Aria, reviewed by the AiCanCode.org team · Updated . How we write and check our notes. Spotted an error? Use Give Feedback at the bottom of the page.

Why it matters

Edge computing and distributed cloud are two approaches engineers use to place compute, storage and networking for systems that must meet latency, privacy or bandwidth constraints. An engineer chooses between them when designing real-time control, IoT, AR/VR, content delivery or analytics pipelines to meet response-time, cost and operational requirements.

Key ideas

  • Edge computing: placing compute and storage at or near data sources (devices, gateways, base stations) to minimize latency and bandwidth use; typically operated by the enterprise or telco, not necessarily by a single cloud provider. Use this when sub-10–50 ms response and on-premises data processing are required or when data must not leave the local site.

  • Fog computing: (define before use) an architecture that extends cloud services to the local area network between the cloud and edges; fog nodes are intermediary devices (routers, gateways) that provide storage and compute. Fog sits between edge and cloud; some sources use it synonymously with edge—state the difference in your answer.

  • Latency components: response delay consists of multiple parts: processing delay (compute time at device/server), transmission delay (time to push bits onto the link), propagation delay (signal travel time over medium), and queuing delay (waiting in buffers). Exact definition: Response time = processing + transmission + propagation + queuing.

  • Bandwidth vs latency trade-off: higher bandwidth reduces transmission time for large payloads; latency (propagation) depends on physical distance and medium speed and cannot be reduced below physics limits.

  • Placement problem: formally, given a set of services, data sources, compute capacities and SLAs, decide where to place services (edge, fog, distributed-cloud region, central cloud) to optimize cost or latency. This is often NP-hard (reduction to facility location). University questions expect you to state assumptions (discrete sites, fixed capacities, latency constraints) and write objective functions (minimize total cost subject to capacity and latency constraints).

  • Consistency and state: moving state to the edge increases the chance of divergence and complicates consistency. Define strong consistency (all clients see same data immediately) and eventual consistency (updates propagate and system converges). Edge systems often accept weaker consistency for availability and low latency.

  • Security and privacy: edge reduces data traversing the core network but increases the attack surface (many small nodes) and operational complexity; distributed cloud centralises management, simplifying uniform security policies but may require data to cross borders (privacy/legal issues).

  • Diagrams to draw in an exam answer (describe exactly):

    1. Draw a layered diagram: bottom layer "Devices / Sensors / Users" with arrows to a middle layer "Edge nodes / Gateways / Base stations" and arrows from that to "Regional distributed cloud nodes" and finally to "Central cloud region". Label arrows with "latency" (short between devices and edge, longer to clouds) and label nodes with compute capacity (e.g., low at devices, medium at edge, high at cloud).
    2. Draw timing-bar diagram for a single request: show segments for transmission, propagation, processing and queuing; label each with formula (transmission = size / bandwidth, propagation = distance / speed).

Formulas

T_response = T_processing + T_transmission + T_propagation + T_queuing — T_response: total response time (s), T_processing: processing time at CPU (s), T_transmission: size / bandwidth (s), T_propagation: distance / v (s), T_queuing: time waiting in buffers (s). Applies to one-way; round-trip can be approximated as 2× when request/response sizes and paths are symmetric; otherwise sum per-direction terms.

T_transmission = S / B — S: message size (bits), B: link bandwidth (bits per second, bps), units: s; applies when full message goes over a link.

T_propagation = d / v — d: distance (m), v: signal speed in medium (m/s), units: s; applies for propagation delay.

Utilization ρ = λ / μ — for a single-server queue (M/M/1) where λ is arrival rate (requests/s) and μ is service rate (requests/s); dimensionless, applies when arrivals are Poisson and service times exponential.

T_queue (M/M/1) = ρ / (μ (1 - ρ)) — average waiting time in queue (s); assumes M/M/1 model and ρ < 1 (stability). Total time in system W = 1 / (μ - λ).

When reasoning about compute times: T_processing = I / P — I: number of instructions or cycles required, P: processing speed (instructions/s or cycles/s), units: s.

Worked examples

  1. Latency comparison (numerical, step-by-step) Given: message size 10 KB (10 × 1024 bytes = 10240 B = 81920 bits), bandwidth B = 10 Mbps (10 × 10^6 b/s), signal speed v = 2.0 × 10^8 m/s (typical fiber estimate), edge distance d_edge = 10 km = 10 000 m, cloud distance d_cloud = 2000 km = 2 000 000 m. Compute one-way and round-trip times. Steps:
  • T_transmission = S / B = 81920 / 10e6 = 0.008192 s = 8.192 ms.
  • T_propagation_edge = d_edge / v = 10000 / 2e8 = 0.00005 s = 0.05 ms.
  • One-way edge = processing (assume negligible) + transmission + propagation ≈ 8.242 ms. Round-trip edge ≈ 16.484 ms. Write Edge RTT ≈ 16.484 ms.
  • T_propagation_cloud = 2000000 / 2e8 = 0.01 s = 10 ms.
  • One-way cloud ≈ 8.192 + 10 = 18.192 ms. Round-trip cloud ≈ 36.384 ms. Write Cloud RTT ≈ 36.384 ms. Conclusion: the edge reduces RTT by roughly 20 ms here — important for interactive applications.
  1. Offloading decision example (numerical) Given a task requiring I = 2 × 10^9 instructions. Device speed P_device = 0.5 × 10^9 instr/s (500 MIPS). Edge server speed P_edge = 20 × 10^9 instr/s (20 GIPS). Data to send S = 1 MB = 1 × 10^6 bytes = 8 × 10^6 bits. Bandwidth B = 10 Mbps = 10 × 10^6 b/s. Assume result size small and ignore return cost; propagation negligible. Steps:
  • Local processing time: T_local = I / P_device = 2e9 / 0.5e9 = 4 s.
  • Transmission time to edge: T_trans = S / B = 8e6 / 10e6 = 0.8 s.
  • Remote processing time: T_remote_proc = I / P_edge = 2e9 / 20e9 = 0.1 s.
  • Total offload time ≈ 0.8 + 0.1 + (small return) ≈ 0.9 s. Offloading is beneficial (0.9 s vs 4 s). Note: If bandwidth were much lower or data larger, offloading could be worse — show this trade-off in answers.
  1. Model-answer long question (theory) Question: "Compare edge computing and distributed cloud for an IoT traffic-monitoring system and state placement strategy." Model answer should: define both terms, give latency and bandwidth calculations, discuss data volumes, state/consistency trade-offs, security/privacy and operational costs, propose a placement: perform preprocessing (filtering, anomaly detection) at edge gateways, aggregate regional analytics at distributed-cloud nodes, central model training in central cloud; justify with latency numbers and data reduction percentages.

Common mistakes

  • Using bytes vs bits inconsistently or mixing decimal (MB = 10^6 B) vs binary (MiB = 2^20 B) units. State which you use and always convert bytes to bits: 1 B = 8 b.
  • Ignoring queuing effects under high load; using only propagation and transmission under realistic loads can under-estimate latency. Remember ρ < 1 for stability and that queueing delay grows rapidly as ρ → 1.
  • Assuming strong consistency is free at the edge; forgetting replication and synchronization costs.

For your exam

Question types you will see: definitions (short), compare-and-contrast (short notes), derivations (latency formulas, queue formulas), numericals (transmission/propagation/RTT, offloading decisions), diagrams (architecture and timing bars), and policy/placement design (model answer). Full-marks answers include: clear definitions (bold terms), assumptions listed, step-by-step derivations or calculations with units, diagrams labelled as described above, and a short justification of design choices including trade-offs.

Quick check

  1. Define edge computing in one sentence.
  2. Give the formula for transmission time and state units.
  3. If a 2 MB message is sent over a 20 Mbps link, what is the transmission time (ms)?

Answers: 1) Edge computing places compute and storage near data sources to reduce latency and conserve bandwidth. 2) T_transmission = S / B where S is bits and B is bits per second, units are seconds. 3) 2 MB = 2 × 10^6 bytes = 16 × 10^6 bits; T = 16e6 / 20e6 = 0.8 s = 800 ms.

Finished this topic? Mark it so your progress, study plan and readiness keep up.

Stuck on something here?