Analytics & ML · Study deck

Practical Edge Computing: Bandwidth and Inference Budgets

Five hundred vibration sensors can produce about 1 MB of samples each second.

Data Dora is your guide for this deck.

edgepatternspractical

Data Dora, the module guide, in a scene from this chapter.

After studying this chapter

Learning objectives

You will be able to:

Test putting numbers to it with a concrete scenario and pass criteria.
Define related chapters and resources with explicit inputs, errors, and change rules.
Explain: Five hundred vibration sensors can produce about 1 MB of samples each second.
Explain: Scenario: A drone delivery company needs to deploy a landing zone detection model on their drones.

Major section

Start With the Decision · Putting Numbers to It

Five hundred vibration sensors can produce about 1 MB of samples each second.

An edge device works near them.
With an illustrative anomaly filter that transmits 5% of the raw-data volume, monthly traffic is about 121 GiB instead of 2,414 GiB; actual reductions depend on the event rate.

Major section

Drone Landing Edge ML Budget · Multi-Model Edge Memory

Scenario: A drone delivery company needs to deploy a landing zone detection model on their drones.

The model must run on-device to ensure safe landing even when cellular connectivity is lost.
They need to calculate if edge inference meets the strict latency requirements.

Major section

Edge ML Latency Calculator · Deployment Patterns and Standards

Industrial sites often expect OPC UA, Modbus, BACnet, IEC 61499, or LWM2M support; cloud-managed edge fleets may use AWS IoT Greengrass, Azure IoT Edge, EdgeX Foundry, or Kubernetes-at-edge tooling.

The right answer is the smallest stack that can meet latency, offline, security, and maintainability requirements without creating a second unmanaged platform.

Major section

Pitfall: Over-Processing at the Edge

Start with simple threshold-based filtering (if temp > 80C, alert).

Measure the false positive/negative rate.
Only upgrade to ML when simple rules prove inadequate.
Always A/B test: deploy ML on 10% of devices, compare accuracy and resource usage.
Set explicit power/compute budgets before selecting algorithms.

Major section

No OTA Update Strategy · Edge Is Not Always Disconnected

The mistake: Deploying edge devices with firmware/software that cannot be remotely updated, leaving the fleet stuck on buggy or insecure versions, requiring expensive truck rolls to fix issues that could be patched remotely.

The fix: Design for the spectrum of connectivity states, not just offline.

Major section

Summary · Concept Check

Factory monitoring example shows edge-hybrid architecture saves 41% over 3 years while meeting 50 ms latency requirements.

Memory optimization example shows 64% savings through activation sharing and INT8 quantization.

Major section

Quick Knowledge Check

For low-volume deployments (<1 GB/day), cloud-only is typically cheaper.

Q: Why can't you rely on average latency alone when designing edge systems?: Cloud average latency might be 200 ms, but peak spikes can reach 2+ seconds during network congestion.
Safety-critical systems must guarantee worst-case latency, not average latency.

Major section

Try It Yourself · Calculate Edge ROI

Scenario: You're proposing edge gateways to reduce cloud costs for a sensor deployment.

Current cloud costs ($/month): _______.
Data reduction factor is the key variable.

Downsample: 60/hour to 1/hour = 60x reduction

Aggregate: 10 sensors to 1 summary = 10x reduction

Filter: Remove 80% unchanged readings = 5x reduction

Total reduction: 60 x 10 x 5 = 3,000x

Major section

Key Takeaway · Continue Your Route

Edge computing decisions should be driven by numbers, not assumptions.

Hybrid architectures -- edge for time-critical decisions, cloud for complex analytics -- typically deliver the best results, and designing for device failures, clock drift, and OTA updates from day one prevents costly rework later.

Deck summary

Key takeaways

Five hundred vibration sensors can produce about 1 MB of samples each second.

Scenario: A drone delivery company needs to deploy a landing zone detection model on their drones.
Industrial sites often expect OPC UA, Modbus, BACnet, IEC 61499, or LWM2M support; cloud-managed edge fleets may use AWS IoT Greengrass, Azure IoT Edge, EdgeX Foundry, or Kubernetes-at-edge tooling.
For a battery-constrained example, a 95% accurate simple filter running 10x longer on battery may beat 99% ML that drains devices in days when both meet the application’s error limits.
For low-volume deployments (<1 GB/day), cloud-only is typically cheaper.

Retrieval practice

Recall check 1 of 5

Data Dora says: answer from memory, then check your reasoning.

Q1What packages code, dependencies and version metadata as one verified installation unit?

AHardware Resource Budget

BSite Network and Device Configuration

CHealth Telemetry and Rollback

DSigned Workload Bundle

Show answer

Answer: D Code, dependencies, and version metadata travel as one verified unit so installation does not mix untrusted or incompatible parts.

Q2Complete the edge computing data aggregation function:

Aself.buffer = deque(maxlen=window_size)

Bself.buffer = list()

Cself.buffer = np.zeros(window_size)

Dself.buffer = deque()

Show answer

Answer: A Edge aggregation uses a fixed-size deque buffer to maintain a sliding window of readings.

Retrieval practice

Recall check 2 of 5

Data Dora says: answer from memory, then check your reasoning.

Q3A factory monitors 200 vibration sensors. Emergency shutdown must trigger within 20 ms; tests show the local machine controller meets that deadline, while the available fog, SCADA, and WAN/cloud paths do not. Weekly trend reports and fleet model training use shared cloud history. Where should each function primarily run?

AAll processing in the cloud; edge devices simply forward raw data

BThreshold checks and shutdown logic at the edge; aggregation and model training in the cloud

CAll processing on an on-premises fog cluster; edge devices act as dumb sensors

DOnly in a centralized SCADA server; no edge or cloud components are needed

Show answer

Answer: B The edge compute patterns chapter emphasises that time-critical decisions (tens of milliseconds) must be made close to the source.

Retrieval practice

Recall check 3 of 5

Data Dora says: answer from memory, then check your reasoning.

Q4A remote oil pipeline uses battery-powered pressure sensors on satellite links that can be offline for hours. The sensors cannot replay missed readings. The gateway must preserve every timestamped reading, raise a local leak alarm even offline, forward remote alarms when connected, and upload the complete record after reconnect. Which edge pattern fits?

ACloud-only processing: buffer nothing locally, then recompute from live telemetry after each satellite outage

BStore-and-forward edge node: local buffering plus alarm logic, with delayed upload when connectivity returns

CStateless edge filtering: drop all readings below threshold and never store history for slow leak review

DAlways-on video analytics at the edge, streaming compressed frames over the intermittent satellite link

Show answer

Answer: B The stated contract requires local alarm logic plus a persistent store-and-forward buffer large enough for the outage.

Retrieval practice

Recall check 4 of 5

Data Dora says: answer from memory, then check your reasoning.

Q5In the seven-level IoT reference model, which level primarily converts data in motion into data at rest (e.g., persistent storage)?

ALevel 1: Physical Devices

BLevel 4: Data Accumulation

CLevel 2: Connectivity

DLevel 3: Edge/Fog Computing

Show answer

Answer: B Level 4 (Data Accumulation) is where data transitions from in-motion to at-rest.

Retrieval practice

Recall check 5 of 5

Data Dora says: answer from memory, then check your reasoning.

Q6In edge processing, what does distillation usually mean?

AEncrypting all sensor data at rest while keeping the same raw sample volume and feature payload unchanged

BConverting IPv4 packets to IPv6 packets while forwarding the complete sensor payload without summarising it

CReducing raw data volume via aggregation/feature extraction while retaining decision-relevant information

DIncreasing sampling rate to capture more detail, then sending every raw sample upstream for later cloud filtering

Show answer

Answer: C Distillation reduces bandwidth and storage costs (e.g., summary stats, FFT bins, anomaly features) while keeping the information needed for local decisions or cloud analytics.

Print reference

Answers 1 of 2

Answer key.

D · Code, dependencies, and version metadata travel as one verified unit so installation does not mix untrusted or incompatible parts.
A · Edge aggregation uses a fixed-size deque buffer to maintain a sliding window of readings.
B · The edge compute patterns chapter emphasises that time-critical decisions (tens of milliseconds) must be made close to the source.
B · The stated contract requires local alarm logic plus a persistent store-and-forward buffer large enough for the outage.

Print reference

Answers 2 of 2

Answer key.

B · Level 4 (Data Accumulation) is where data transitions from in-motion to at-rest.
C · Distillation reduces bandwidth and storage costs (e.g., summary stats, FFT bins, anomaly features) while keeping the information needed for local decisions or cloud analytics.