1.5 Evolving Cloud Technologies: AI/ML Services, IoT & Edge Computing

Evolving Cloud Technologies: AI/ML Services, IoT & Edge Computing

Objective 1.11 of the Cloud+ (CV0-004) blueprint requires candidates to identify evolving technologies built on cloud platforms — not to operate them. The exam tests whether you can match a business requirement (transcribe a support call, read a scanned invoice, stream factory telemetry) to the correct managed service category or architecture. Edge computing also appears under Objective 1.2 (service availability), because edge locations push cloud services physically closer to end users and devices.

---

1. Cloud-Managed AI & Machine Learning Services

Every hyperscaler delivers machine learning as consumption-based APIs. Instead of collecting and labeling training data and operating GPU clusters, an application simply calls a pre-trained inference endpoint and pays per request.

Key Concepts Behind the Services

• Inference vs. training: The services above perform inference — applying a model the provider already trained. Building a custom model on proprietary data belongs to the training platforms (Amazon SageMaker, Azure Machine Learning, Vertex AI), which require labeled datasets and GPU capacity provisioning.
• Generative AI: Foundation-model platforms (Bedrock, Azure OpenAI, Vertex AI) bill per token and raise distinct governance questions: prompts may contain PII, outputs require grounding techniques such as Retrieval-Augmented Generation (RAG) against enterprise data, and many organizations route access through a private API gateway with content filters.
• Exam association table: read text from scanned receipts → OCR (Textract); decide whether product reviews are angry → sentiment analysis (Comprehend); a contact-center IVR speaks the account balance → text-to-speech (Polly); transcribe recorded support calls → speech-to-text (Transcribe); a chatbot that drafts answers → generative AI (Bedrock).

[!IMPORTANT] Exam Trap: "Extract printed tracking numbers from two million scanned shipping labels" is an OCR / text-recognition requirement. Do not choose visual object detection (which identifies things in a picture, not characters) or generative AI (which composes new text rather than reading it).

---

2. Internet of Things (IoT) Cloud Architecture

An IoT estate moves telemetry through four layers, each of which the exam expects you to identify:

+---------------------------------------------------------------------------------------------+
|                              FOUR-LAYER IOT CLOUD ARCHITECTURE                              |
|                                                                                             |
|   [ 1. SENSORS + ACTUATORS ]        [ 2. EDGE GATEWAY ]        [ 3. INGESTION BROKER ]      |
|   - Measure temperature, GPS,       - Protocol translation     - Authenticates fleets of    |
|     vibration, flow; actuators      - X.509 device certs       - devices                    |
|     drive valves, relays, locks     - Store-and-forward        - AWS IoT Core (device       |
|   - Battery-powered, constrained      buffering during WAN     -   shadows + rules engine)  |
|     compute/memory                    outages; local filter    - Azure IoT Hub (device      |
|                                        and pre-aggregation       twins)                   |
|              |                             |                             |                  |
|              +---- MQTT / CoAP / LPWAN --->+---- MQTT over TLS (8883) ----+                  |
|                                                                                             |
|                                       [ 4. CLOUD PROCESSING ]                               |
|                                       - Streaming analytics, dashboards, ML inference,      |
|                                         long-term data-lake storage                         |
+---------------------------------------------------------------------------------------------+

Communication & Transmission Protocols

• MQTT (Message Queuing Telemetry Transport): The default IoT protocol. A lightweight publish/subscribe design over TCP with TLS (port 8883) in which devices publish to topics on a central broker; Quality-of-Service levels offer QoS 0 (fire-and-forget), QoS 1 (at-least-once delivery), and QoS 2 (exactly-once). Ideal for low-power devices on unreliable links.
• CoAP (Constrained Application Protocol): A UDP-based, REST-like protocol for severely constrained devices that cannot afford TCP overhead.
• HTTP(S): Acceptable for mains-powered devices with generous bandwidth, but its per-request overhead is heavy for battery telemetry.
• Edge radio links: LPWAN options (LoRaWAN, NB-IoT) carry long-range, low-bitrate sensor traffic; BLE and Zigbee serve short-range personal-area networks that hand off to a gateway.

[!NOTE] Exam associations: unreliable backhaul + minimal power draw → MQTT; the cloud component that authenticates devices and brokers their messages is the IoT platform (AWS IoT Core / Azure IoT Hub); when connectivity drops, the edge gateway buffers telemetry (store-and-forward) and synchronizes on reconnection.

---

3. Edge Computing Architecture

Edge computing executes workloads physically close to the data source rather than in a distant cloud region:

• Why edge: round-trip latency (carrier-embedded compute can reach single-digit milliseconds versus 50–150 ms to a remote region), backhaul bandwidth (process factory video locally and ship only anomalies), intermittent connectivity, and data-sovereignty constraints that keep raw data on premises.
• The edge spectrum:
  1. CDN points of presence (PoPs): CloudFront, Azure Front Door, and Cloudflare cache static assets and run lightweight edge functions near users.
  2. Provider edge compute: AWS Wavelength embeds compute inside carrier 5G networks; AWS Local Zones and Azure Extended Zones place compute in metro areas.
  3. On-premises edge hardware: AWS Outposts racks, Azure Stack Hub/Edge, and Google Distributed Cloud run provider-native services inside the customer's facility.
• Availability relationship (Objective 1.2): Edge PoPs raise service availability for regional users, and local edge processing lets sites keep operating — running control loops and buffering data store-and-forward — while the WAN link is down.
• Management model: Edge nodes are centrally orchestrated from the cloud but tolerate disconnection. Over-the-air software updates roll out in deployment rings with automatic rollback on health-check failure.

[!CAUTION] Exam Trap: "Users in one metro area need sub-10 ms latency for an augmented-reality application" points to Local Zones or carrier-edge (Wavelength) compute — not simply choosing a different full cloud region. "The plant must keep running when disconnected from the internet" points to an IoT edge gateway with local compute and store-and-forward buffering.

---

4. Exam-Ready Distinctions