Performance Monitoring, Workload Planning, and Capacity Management
Performance and capacity questions test whether you can read the right metric, put it in the right unit, and forecast with it. Unisphere for PowerMax collects the data; Workload Planner and SLO compliance turn it into decisions.
Metrics
- IOPS, throughput (MB/s), response time (ms), queue depth and utilization (%) are reported per array, storage group, device, front-end port and back-end component. Response time is what hosts feel; utilization shows how busy a component is.
- Real-time charts answer what is happening now; historical data answers what changed at 14:00 and whether it is a trend.
- SLO compliance compares each storage group's measured response time with its SLO target, so a generally healthy array can still have one unhappy storage group.
Capacity
- SRP capacity is consumed by data, snapshot and metadata separately. Provisioned (subscribed) capacity is what thin devices claim; consumed capacity is what has actually been written.
- Subscription ratio is provisioned divided by usable. Over-subscription is normal for thin volumes but must be justified against growth and snapshot usage.
- Effective capacity applies data reduction (compression and, where available, deduplication) to estimate usable capacity; never mix effective and raw numbers in one report.
Workload Planner
Workload Planner models an array from collected performance data; you add hypothetical devices or workloads and it forecasts response time and utilization. It is a what-if tool, not a live view of the array.
Worked example
An SRP provides 100 TB usable, 300 TB are provisioned and the measured data reduction ratio is 3:1. Subscription is 300%, which looks alarming, but effective capacity is roughly 300 TB, about level with provisioned. Before adding 50 TB, check consumed rather than provisioned capacity, snapshot consumption and peak port utilization, then model the addition in Workload Planner instead of trusting the ratio alone.
Traps
Reporting MB/s as IOPS; reading array-level averages when a single storage group is skewed; treating provisioned capacity as consumed; and assuming a data reduction ratio applies to data that is already compressed or encrypted.