Lesson 0001 · DTS 208

Cloud, Fog, and Edge Architectures

The three-tier computing hierarchy, core motivations for edge computing, service and deployment models, and workload partitioning.

§1 The Evolution of Computing Architectures

Historically, computing progressed from centralized mainframes to decentralized PCs, and then to centralized Cloud Computing. While cloud computing offers virtually unlimited storage and massive parallel processing, the explosion of Internet of Things (IoT) sensors, autonomous vehicles, and industrial machinery created critical bottlenecks: network latency, bandwidth congestion, privacy risks, and reliance on continuous internet connectivity.

To resolve these challenges, modern distributed systems deploy a three-tier architecture:

The Three-Tier Hierarchy

1. Cloud Layer (Centralized): Massive hyperscale data centers (AWS, Azure, Google Cloud). Responsible for heavy model training, global data aggregation, historical analytics, and long-term archiving. (Latency: 50–200ms+).
2. Fog Layer (Intermediate / LAN): Local network gateways, micro-data centers, or factory servers. Provides localized compute, data aggregation, protocol translation, and regional caching. (Latency: 10–50ms).
3. Edge Layer (Decentralized / Source): Devices and microcontrollers directly at or near the physical data source (smartphones, IoT gateways, smart cameras, PLCs). Performs real-time inference, immediate filtering, and local control. (Latency: <1–10ms).

§2 Core Drivers: Why Move to the Edge?

Exam questions in DTS 208 regularly test the four fundamental drivers of Edge Computing:

Driver	Problem in the Cloud	Edge Computing Solution
1. Latency	Round-trip time across the WAN takes 50–500ms; unacceptable for autonomous braking or robotic safety stops.	Near-instantaneous local inference in sub-millisecond to 10ms timeframe.
2. Bandwidth	Streaming raw HD video feeds or high-frequency vibration sensors saturated network uplinks and incurs high cloud ingress/storage costs.	Data is processed and filtered locally; only high-level events, anomalies, or summaries are transmitted upstream.
3. Privacy & Security	Transmitting raw personal video, medical vitals, or trade secrets exposes sensitive data in transit.	Raw data remains on-device; only anonymized telemetry or model updates leave the local perimeter.
4. Reliability / Offline Operation	Cloud dependence means factory floor stops if internet connection drops.	Edge devices operate autonomously 24/7 without active internet connections.

§3 Cloud Service & Deployment Models

Service Models

IaaS (Infrastructure as a Service): Cloud provider delivers raw virtual machines, storage, and networking (e.g. AWS EC2, Google Compute Engine). User manages OS, runtime, and application.
PaaS (Platform as a Service): Cloud provider manages OS, runtime, and hardware; user deploys application code (e.g. AWS Elastic Beanstalk, Heroku).
SaaS (Software as a Service): End-user software hosted entirely by provider (e.g. Google Docs, Salesforce, Microsoft 365).

§4 Workload Partitioning: What Stays vs. What Goes

Modern distributed systems partition workloads according to resource demands:

Keep at the Edge: Real-time obstacle detection, anomaly alerts, sensor filtering, private facial authentication, local closed-loop robotic feedback.
Send to the Cloud: Training deep learning models, multi-factory trend analysis, longitudinal billing data, complex global optimization algorithms.

Cloud, Fog, and Edge Architectures

§1 The Evolution of Computing Architectures

§2 Core Drivers: Why Move to the Edge?

§3 Cloud Service & Deployment Models

§4 Workload Partitioning: What Stays vs. What Goes

§5 Check yourself

Sources