How Should 100,000 Mobile App Instances Stream CPU Telemetry into Azure SQL?

Why the Answer Is Correct

Azure Event Hubs is a fully managed, partitioned event ingestion service built for exactly this shape of workload: hundreds of thousands of devices each emitting small telemetry messages. It absorbs the bursty per-instance usage data at low latency and low cost per million events, and it decouples the mobile tier from the analytics tier so a downstream outage never blocks device reporting. Azure Stream Analytics then reads that stream and applies a one-hour tumbling window grouped by the app instance identifier, producing the hourly average CPU utilization with no servers to patch. Stream Analytics natively outputs to Azure SQL Database, so the average is upserted into the relational store where reporting and Power BI can consume it. Together the pair satisfies both explicit requirements (accept data, compute hourly per-instance averages, write to SQL) while honoring the cost and administrative-effort constraint, because both are PaaS with no cluster or VM management.

Why Other Candidate Answers Are Wrong

Writing directly from the mobile app to Azure SQL Database fails on volume and cost: 100,000 devices opening connections and issuing INSERT/UPDATE statements will exhaust DTU/vCore limits, force expensive scale-up, and give you no buffering when the database is busy. Azure Functions triggered per message is technically possible but is a poor fit for hourly windowed aggregation, since you must persist per-instance state in a durable store and pay for millions of executions. A batch alternative such as Azure Data Factory or Blob Storage plus a scheduled job adds latency of hours and requires orchestration, which contradicts "minimizes administrative effort." Azure IoT Hub would work for device messaging but is priced and licensed per-unit around device management features that a plain mobile telemetry scenario does not need, and it still requires a separate stream processor to compute the averages. Logic Apps and Event Grid are not designed for high-throughput streaming aggregation at all.

Community Comment Notes

AllanHouston argued for Azure Event Hubs on the grounds that it is a "fully managed, low-latency data ingestion service" engineered for telemetry and high-throughput messaging, and added that it "easily absorbs millions of events per second" with minimal cost and administrative setup compared with a database. That reasoning matches the exam's intent: the ingestion option must be a buffered, partitioned event service, not a relational store. He then continued that the averaging and the write to the database are handled by the downstream processing service, which is precisely the two-step split this question rewards. Treat the vote tally as a hint only — the justification, not the count, is what should drive your selection.