Skip to content
← All resources

Intermediate · 8 lessons · Modernize the Data Estate

Fabric Capacity and Cost Engineering

Run a Fabric capacity on evidence: diagnose throttling, cut the cost of the work, size and protect the capacity deliberately, and give finance a bill they can explain.

Most Fabric cost problems do not start with the price list. They start with a capacity that nobody is watching, a refresh that runs hourly for data that changes daily, and a throttling incident that ends with someone approving a bigger SKU because it was the fastest thing to approve.

A bigger SKU gives inefficient work more room to run. It does not make that work efficient, and it does not tell you why the capacity was under pressure.

This course treats capacity the way a good engineering team treats any shared system. You will learn the billing model well enough to predict it, then read the evidence in the Capacity Metrics app, diagnose incidents by their shape, and fix the workload where the compute was actually consumed. Only then do we size, protect, and scale the capacity, because by that point you will know what legitimate demand looks like.

The final lessons move from engineering to operations. Reservations, pausing, and scheduled resizing turn a usage pattern into a cost plan. Chargeback, alerts, and a short weekly review keep that plan honest after the project team moves on.

Work through the lessons in order and complete each exercise against your own capacity where you can. The numbered examples in this course are illustrative unless they quote Microsoft documentation, so replace them with your own measurements as you go. By the end you will have a baseline, an incident worksheet, a sizing envelope, a guardrail decision, a cost plan, and a one-page operating runbook.

Lessons

  1. 1 How Fabric Capacity Really Bills Capacity units, CU-seconds, smoothing, bursting, and carryforward, plus the meters that sit outside the capacity, so you can predict the bill before you try to control it. 10 min
  2. 2 Reading the Capacity Metrics App Install the Capacity Metrics app, learn the handful of visuals that matter, and record a baseline before you change anything. 10 min
  3. 3 Diagnosing Throttling From the Capacity Down Confirm whether the capacity actually throttled, find the operations that created the pressure, and classify the incident before anyone proposes a fix. 11 min
  4. 4 Making the Work Cheaper Reduce CU-seconds per operation and operations per day across warehouses, Spark, pipelines, semantic models, and Real-Time Intelligence, then prove the saving with before-and-after evidence. 12 min
  5. 5 Sizing a Capacity With Evidence Turn a baseline, an estimate, and a pilot into a capacity envelope with stated assumptions, including the demand that AI and data agents add. 11 min
  6. 6 Guardrails: Surge Protection, Overage, and Isolation Protect critical work when demand runs ahead of the capacity, and choose deliberately between surge protection, capacity overage, separate capacities, and on-demand billing for Spark. 11 min
  7. 7 Scaling, Pausing, and Reservations Match what you pay to the shape of your usage: reserve the steady baseline, scale for predictable peaks, pause what does not need to run, and automate it safely. 11 min
  8. 8 A Capacity Operating Model Give every consumer an owner, show each team what it uses, alert before users notice, review on a schedule, and keep the whole course on one page as a runbook. 10 min