DP-900 · 4 Describe an analytics workload
Batch versus streaming data
Exam objective: Describe the difference between batch and streaming data
Batch processing groups records and processes them together. Stream processing handles each event as it arrives, with much lower latency.
Data processing turns raw records into something useful, and there are two timing models for doing it.
Batch processing collects records first and processes the whole group together, for example once an hour, once a certain volume has arrived, or on some other trigger. A monthly credit card statement is a batch result: every purchase from the month is rolled into one bill rather than billed as it happens.
Stream processing handles each event as it arrives, continuously and in near real time. A smoke detector has to work this way: waiting for a batch to accumulate before raising the alarm would defeat the point.
| Aspect | Batch | Streaming |
|---|---|---|
| Data scope | The whole dataset | Recent data or a time window |
| Typical latency | Hours | Seconds or milliseconds |
| Typical analysis | Complex, exploratory | Simple aggregates, rolling calculations |
Many solutions use both at once. A lambda architecture captures events in real time for a live dashboard, while also writing the same raw events to a data store so they can be reprocessed later alongside historical, batch loaded data.
On the exam, words like "scheduled", "once a threshold is reached" or "a monthly report" signal batch, while "as it happens", "continuously" or "immediate alert" signal streaming.
Key points
- Batch processing collects records and processes the whole group together, on a schedule or other trigger.
- Stream processing handles each event individually, continuously, as it arrives.
- Batch processing can see the whole dataset; streaming typically only sees recent data or a rolling window.
- Streaming has much lower latency than batch, often seconds or milliseconds instead of hours.
- A lambda architecture combines both: real time processing for live views and batch processing for historical analysis of the same data.
Exam trap
Streaming is not automatically the right choice just because data arrives continuously. If a scenario needs complex analysis over the full dataset rather than an immediate response, batch is still the better fit.
Check yourself
For each statement about batch and stream processing, select Yes if it is true. Otherwise, select No.
- Batch processing typically has latency measured in hours, while stream processing typically has latency in seconds or milliseconds.
- Stream processing usually has access to the full historical dataset at all times.
- Batch processing is well suited to complex analytics over large datasets.
Go deeper on Microsoft Learn
Checked against Microsoft Learn on October 1, 2026.