You are managing a Microsoft Fabric Capacity. You need to identify which two metrics are most effective for tracking the 'smoothing' behavior of your capacity during peak usage. Which two metrics should you monitor?
Trap 1: Data storage growth rate
Data storage growth rate measures the volume of data added to OneLake over time. While important for cost management and storage planning, it has no direct relationship to the compute smoothing mechanism, which is exclusively concerned with CPU and memory utilization during active query and pipeline execution.
Trap 2: Network latency between regions
Network latency between regions is an infrastructure metric that is generally outside the control of capacity management. It does not provide actionable insights into how Fabric's compute smoothing handles bursty workloads, making it irrelevant for monitoring the effectiveness of internal capacity resource allocation and load balancing.
Trap 3: Workspace file count
Workspace file count measures the number of objects stored in the environment. This metric is useful for auditing and storage optimization, but it does not track compute performance or the smoothing logic applied to Spark jobs or SQL queries executed within the Fabric capacity environment.
- A
Capacity Utilization percentage
Capacity utilization represents the actual compute usage against your SKU. Tracking this metric allows you to visualize how smoothing impacts the overall load, showing you if your workloads are consistently peaking or if they remain within the sustained performance thresholds set by your current capacity tier.
- B
Throttling Events
Throttling events occur when the smoothing mechanism cannot fully compensate for sustained demand exceeding the capacity limits. Monitoring these events is essential to determine if your current SKU is under-provisioned, as it directly indicates where smoothing has failed to prevent performance degradation for the users involved.
- C
Data storage growth rate
Why it fails: Data storage growth rate measures the volume of data added to OneLake over time. While important for cost management and storage planning, it has no direct relationship to the compute smoothing mechanism, which is exclusively concerned with CPU and memory utilization during active query and pipeline execution.
- D
Network latency between regions
Why it fails: Network latency between regions is an infrastructure metric that is generally outside the control of capacity management. It does not provide actionable insights into how Fabric's compute smoothing handles bursty workloads, making it irrelevant for monitoring the effectiveness of internal capacity resource allocation and load balancing.
- E
Workspace file count
Why it fails: Workspace file count measures the number of objects stored in the environment. This metric is useful for auditing and storage optimization, but it does not track compute performance or the smoothing logic applied to Spark jobs or SQL queries executed within the Fabric capacity environment.