Skip to content

An independent study reference written by Dr Phuc V. Nguyen. It is not official subject material — for assessment requirements always follow your subject outline and vUWS.

Cloud for analytics

Cloud computing supplies storage and processing as a metered service you request and release without buying hardware. The widely used definition names five essential characteristics, including on-demand self-service, rapid elasticity and measured service, and three service models: infrastructure, platform and software as a service. For analytics the decisive change is the separation of storage from compute. Data can be kept cheaply and permanently while processing capacity is created for a query and destroyed afterwards. That removes the old constraint of sizing a server for the busiest hour of the year, and replaces it with a bill that moves with what you run.

Why it matters

Owning a server is like owning a van you need twice a month. It costs the same whether you drive it or not, and on the one day you need three vans you are stuck. Cloud capacity is hiring the van by the hour, which suits uneven work, and analytics is uneven work. A warehouse sits idle all night and then twenty people open dashboards at nine. The catch is the meter. Nobody leaves a hired van running for seven days by accident. A scheduled query does exactly that.

Before you read on — recall

A warehouse is billed on the volume of data each query scans. Which change most reliably reduces the bill without changing what users see?

Formulas

Query cost under on-demand, scan-based pricing
C=q×v×pC = q \times v \times p
Where qq is the number of runs, vv the volume each run scans and pp the price per unit scanned. A dashboard query scanning 4 TB and refreshing hourly scans about 2,900 TB a month. The same query restricted to one date partition, scanning 20 GB, scans about 14 TB. The query text barely changes and the bill changes by a factor of two hundred. This arithmetic belongs to one billing model, the on-demand tier of a scan-priced service such as BigQuery. Platforms that sell provisioned capacity or compute credits charge for the time a cluster is available or running, so there the levers are idle capacity, cluster size and concurrency rather than bytes scanned.

Worked examples

Scenario

A team migrates its warehouse to a cloud platform and the monthly bill triples in the second month, with no increase in users.

Solution

Elastic pricing removed the discipline a fixed server used to impose. Three causes usually explain it. Dashboards set to refresh every fifteen minutes run whether anyone is watching or not. Queries select every column from a wide table, because at small scale there was no penalty. And tables are unpartitioned, so a filter on the last seven days still scans four years. None of these are faults of the platform. They are habits that a fixed-capacity machine punished with slowness and a metered service punishes with an invoice.

Scenario

An Australian health services provider wants to use an overseas cloud region because the unit price is lower.

Solution

Price is not the only variable. Under the Australian Privacy Principles, an entity that discloses personal information to an overseas recipient must take reasonable steps to ensure the recipient does not breach those principles, and it generally remains accountable for what happens to the information. Some health datasets carry stricter location requirements again. The workable answer is usually to keep identified data in an Australian region and, where an overseas service is genuinely needed, send de-identified or aggregated extracts instead.

Common mistakes

  • Moving to the cloud reduces cost automatically. It converts a fixed cost into a variable one, and a variable cost falls only if usage is managed; a badly written or over-scheduled query costs more in the cloud than on a server already paid for.
  • The cloud provider is responsible for the data. Responsibility is shared, and the dividing line moves with the service model, so it has to be read from the specific agreement rather than assumed; across all three models, classification, access rights, retention and lawful handling of personal information stay with the organisation.
  • Software as a service means there is nothing left to manage. The infrastructure is managed for you; definitions, access, integrations and data quality are not, and those are where analytics work actually fails.
  • Elastic means unlimited and instant. Concurrency caps, quota limits, warm-up time and charges for moving data back out all still apply, and the outward direction is usually the expensive one.

Revision bullets

  • Five essential characteristics include on-demand self-service, rapid elasticity and measured service
  • Service models: infrastructure, platform and software as a service
  • Separating storage from compute is what changed the economics of analytics
  • Under scan-based pricing, partitioning and column pruning cut the bill; under capacity or compute pricing the levers are idle capacity and concurrency
  • Data residency and accountability for personal information stay with the organisation

Quick check

A warehouse is billed on the volume of data each query scans. Which change most reliably reduces the bill without changing what users see?

A vendor says its fully managed analytics platform removes the need for data governance. What is wrong with the claim?

Connected topics

More in Data Foundations

Sources

  1. Mell, P., & Grance, T. The NIST Definition of Cloud Computing. NIST Special Publication 800-145, National Institute of Standards and Technology, 2011.
    The reference definition of the five essential characteristics, three service models and four deployment models.
  2. Armbrust, M., Fox, A., Griffith, R., Joseph, A. D., Katz, R., Konwinski, A., Lee, G., Patterson, D., Rabkin, A., Stoica, I., & Zaharia, M. "A View of Cloud Computing." Communications of the ACM, 53(4), 50-58, 2010.
    Sets out elasticity and pay-as-you-go pricing as the economic change, along with the obstacles including data transfer costs and lock-in.
  3. Office of the Australian Information Commissioner. Australian Privacy Principles, made under the Privacy Act 1988 (Cth).
    APP 8 governs cross-border disclosure of personal information and the steps required of the disclosing entity.
How to cite this page
Dr. Phil's Quant Lab. (2026). Cloud for analytics. Derivatives Atlas. https://phucnguyenvan.com/concept/ba-cloud-for-analytics
Next concept
The data warehouse
Built by Dr. Phuc V. Nguyen ·Follow on LinkedInWork with PhilEmail