Data Storage · Study deck

Database Selection for IoT Workloads

Picture a sensor record that is read now, joined next week, and removed at the end of a job.

Data Dora is your guide for this deck.

databaseselection

After studying this chapter

Learning objectives

You will be able to:

Explain: The registry table's fixed device_id / zone / fw columns make it queryable by a stable schema, so the example *lands in: SQL row store*.
Explain: For 10,000 meters at one reading per minute, the store receives $10{,}000\times1{,}440=14.4$ million readings per day.
Explain: A time-series path can optimise append and time-window queries, while relational work-order data needs joins and transaction rules.
Explain: A database label does not settle the choice; match each store to its role, failure behavior, and recovery record.

Major section

Start With the Next Data Change

One store can make the first build easy, but it may strain under a new query or fault.

More stores fit each job, yet they add copies and repair work.
A database label does not settle the choice; match each store to its role, failure behavior, and recovery record.
Firmware is the code stored on a device.

Major section

Overview: Choose A Database From Evidence

The registry table's fixed device_id / zone / fw columns make it queryable by a stable schema, so the example *lands in: SQL row store*.

Database selection is a release decision, not a popularity contest.
An IoT system may store telemetry, device registry records, events, media artifacts, latest dashboard values, and analytics outputs.
Those records do not all have the same query path, consistency need, retention window, or recovery requirement.

Data storage evidence loop moves from the data contract through ingestion retries and duplicates, validation quality, store role with indexing and partitioning, retention and archiving lifecycle, and restore evidence from replay and audit.

Major section

Practitioner: Build The Storage Role Ledger

A practical selection review separates storage role from product choice.

The role describes the job the data store must do.
The product and deployment shape are selected only after representative workloads prove the role can be operated safely.
Splitting too late can hide lifecycle, query, and consistency problems.

Major section

Under The Hood: Selection Is A Failure-Mode Decision

A database can be suitable for one role and unsafe for another depending on configuration and evidence.

For IoT systems, the same physical device can produce different records with different authority.
A raw temperature measurement may tolerate eventual arrival.
A firmware-assignment record may need tighter authority and audit behavior.

Major section

Place Meter Readings and Work Orders Deliberately

For 10,000 meters at one reading per minute, the store receives $10{,}000\times1{,}440=14.4$ million readings per day.

A time-series path can optimise append and time-window queries, while relational work-order data needs joins and transaction rules.
Using two stores may fit better than forcing both shapes into one database.
Database scale must be checked against the 14.4-million-reading daily workload and failure target.

Structured, semi-structured, and unstructured IoT data, each with a concrete sample and an example store.

Deck summary

Key takeaways

One store can make the first build easy, but it may strain under a new query or fault.

The registry table's fixed device_id / zone / fw columns make it queryable by a stable schema, so the example *lands in: SQL row store*.
A practical selection review separates storage role from product choice.
A database can be suitable for one role and unsafe for another depending on configuration and evidence.
For 10,000 meters at one reading per minute, the store receives $10{,}000\times1{,}440=14.4$ million readings per day.

Retrieval practice

Recall check 1 of 4

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

Q1What is the best first step when selecting storage for a new IoT telemetry and device-management system?

APick whichever database is currently popular for dashboards.

BDescribe each data class, query path, and owner before naming a database product.

CUse a cache as the only source of truth because dashboard latency matters.

DDelay retention, restore, and owner decisions until after production traffic starts.

Show answer

Answer: B Database selection becomes reviewable when the workload and release evidence are visible before product choice.

Retrieval practice

Recall check 2 of 4

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

Q2A team adds a separate media store. The upload demo works, but there is no restore test or owner. What should block release?

ANothing; a successful upload proves recovery.

BOnly a faster dashboard query at launch.

CAn owner and tested restore of media and metadata.

DA second copy in the same untested store.

Show answer

Answer: C The extra store needs an owner and restore proof tied to its data and metadata.

Retrieval practice

Recall check 3 of 4

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

Q3A proposal adds separate stores for telemetry, registry, search, graph, cache, artifacts, and analytics before any representative query evidence exists. What should the review require first?

AApproval based on isolating each workload in a specialized database.

BA role-by-role evidence table for each proposed store.

CSeparate identifiers in each store to support independent schema changes.

DOnly a vendor-logo comparison without query evidence.

Show answer

Answer: B Additional stores can be justified, but every role needs workload evidence and operational ownership.

Retrieval practice

Recall check 4 of 4

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

Q4A latest-value cache makes a dashboard fast, but the durable telemetry store sometimes lags and the cache does not show freshness. What is the strongest review finding?

AThe cache should become the only source of truth because it is faster.

BThe lag is acceptable because fast dashboards outweigh missing history.

CThe dashboard colours alone need to change, not cache freshness.

DThe cache needs a durable source-of-truth link and freshness label.

Show answer

Answer: D Latest values are useful, but users need evidence about freshness and rebuild behavior.

Print reference

Answers

Answer key.

B · Database selection becomes reviewable when the workload and release evidence are visible before product choice.
C · The extra store needs an owner and restore proof tied to its data and metadata.
B · Additional stores can be justified, but every role needs workload evidence and operational ownership.
D · Latest values are useful, but users need evidence about freshness and rebuild behavior.