2.1 Implement and manage an analytics solution Overview

How Microsoft Fabric Is Organized

Microsoft Fabric is a unified, software-as-a-service (SaaS) analytics platform that runs on top of a single logical data lake called OneLake. For DP-700, you must know the object hierarchy cold, because most questions are really asking "at which level is this setting or permission applied?"

The hierarchy from top to bottom is: tenant (your Microsoft Entra organization) holds one or more capacities; capacities power workspaces; workspaces contain items (lakehouses, warehouses, eventhouses, KQL databases, notebooks, pipelines, semantic models, reports); and workspaces are optionally grouped into domains for federated governance.

A key SaaS principle: there are no servers to patch and no clusters to size by hand. You buy a capacity, assign workspaces to it, and Fabric meters the compute the workloads consume.

Capacities and SKUs

A Fabric capacity is a reserved pool of compute measured in Capacity Units (CUs). SKUs are named with an F prefix and the CU/second allowance: F2 = 2 CU/s, F8 = 8, F64 = 64, scaling up to F2048. Capacities are pay-as-you-go: you can scale them up or down, and pause them when idle so you stop being billed for compute (the F SKU is pausable; legacy P SKUs are not).

F64 is the magic threshold. At F64 and above you get full Power BI Premium feature parity, and free (non-Pro) users can view content in those workspaces. Below F64 (F2-F32), viewers still need Power BI Pro licenses, and several features are gated off, including the XMLA read/write endpoint, paginated reports, and email subscriptions.

• Smoothing: Fabric averages bursty CPU usage over a rolling window so short spikes do not immediately throttle you.
• Throttling: when sustained demand exceeds the SKU's CU allowance, Fabric delays interactive operations, then background jobs, then rejects requests.
• Autoscale: an optional safety net that adds CUs on demand. It is cost-effective for rare spikes; if it fires daily, upgrading the SKU is cheaper.
• Bursting: a single job can temporarily use more CUs than the SKU allows to finish faster, repaid by smoothing.

Many exam stems hinge on "viewers report they cannot open content for free" or "this feature is unavailable" — the answer is almost always "the capacity is below F64."

OneLake, Workspaces, and Domains

OneLake is the one logical data lake provisioned automatically for the whole tenant — "OneDrive for data." There is exactly one OneLake per tenant. Every workspace becomes a top-level folder in OneLake, and every data item (lakehouse, warehouse) gets a folder beneath it. All Fabric engines write open Delta Lake / Parquet by default, so the same physical files are readable by every engine without copies. Shortcuts let you reference data in another OneLake location, ADLS Gen2, Amazon S3, or Google Cloud Storage without moving or duplicating it.

Domains are a governance construct: a domain is a named grouping of workspaces (for example, Sales, Finance, HR). A Fabric admin assigns domain admins, who can in turn add or remove workspaces and delegate certain tenant settings down to the domain. Subdomains refine the grouping. Domains drive discoverability in the OneLake catalog and let business units self-govern.

The takeaway for the exam: OneLake is a storage layer (one per tenant), workspaces are the unit of collaboration and ALM binding, and domains are a governance overlay — do not confuse a domain (governance) with a capacity (compute).

Workloads, Items, and Where Compute Comes From

Fabric unifies several workloads under one capacity: Data Engineering (lakehouse, Spark notebooks), Data Warehouse (T-SQL warehouse), Data Factory (pipelines, dataflows), Real-Time Intelligence (eventstreams, eventhouse, KQL), Data Science (ML models), and Power BI (semantic models, reports). Every workload draws compute from the same capacity, so a heavy Spark job and a Power BI refresh compete for the same CU pool. That shared-pool design is why capacity sizing and smoothing matter: a single under-sized F SKU can throttle unrelated teams sharing it.

Each workload produces items. A lakehouse stores files and Delta tables and auto-generates a read-only SQL analytics endpoint plus a default semantic model. A warehouse is a fully transactional T-SQL store. An eventhouse/KQL database handles high-volume telemetry. A pipeline orchestrates; a notebook transforms; a dataflow Gen2 does low-code shaping. Knowing which item produces which capability lets you answer "which item should I create?" questions quickly.

Trial and licensing context

DP-700 also expects light awareness of licensing. A Fabric (Free) license lets a user create items only in a My workspace or a trial. A Power BI Pro or Premium Per User (PPU) license is required to share content when the capacity is below F64. At F64+, content can be shared to free users. The Fabric capacity (the F SKU) is what actually powers non-Power-BI Fabric workloads; without a capacity assigned to a workspace, those Data Engineering and Warehouse items cannot run.

Keep these three things distinct: the per-user license (who can author/share), the capacity (what compute runs the workloads), and the workspace (where items live).