Understanding workload monitoring, planning, and provisioning

Unisphere supports applications that support the monitoring, planning, and provisioning of workloads.

If you have an array that is monitored for performance by Unisphere for PowerMax for at least 2 weeks, then you have enough data to do a characterization of the workload for an existing storage group or multiple storage groups.

The workload planner (WLP) application performs performance impact checks (the terms performance impacts checks, admissibility checks, and suitability checks are used interchangeably) when provisioning using the Unisphere for PowerMax wizards; it performs the checks to ensure that there is sufficient performance headroom before creating application workloads on storage arrays. WLP keeps its own database which is a subset of Key Performance Indicator (KPI) metrics that Unisphere for PowerMax gathers and uses this to help make decisions, so to use it, the arrays in question must be registered for performance data collection.

WLP models the internal load balancing of the array when adding additional load to an existing array. WLP can help to place workloads on the most suitable target array for new provisioning and/or migrated workloads. WLP can help with creating or selecting port groups, based on existing zoning and utilization while balancing across directors.

A Note on Data Processing and Organization

Many applications (and as a result, storage systems) experience fluctuations in workloads over time. Often, these fluctuations repeat in a predictable way. To keep the Workload Planner data footprint as small as possible, but still adequately represent the history and fluctuations in workloads, multiple weeks of historical storage system performance data are weighted and averaged into a week's worth of four-hour time windows which create a "representative workload" for all system components and applications that are applied to a storage system.

Best practice limits are applied to account for maximum performance capabilities of a system component. All performance impact scores and utilizations that are calculated and reported by Workload Planner are done so against these best practice Limits, not the actual maximum performance capacity of the pertinent component or components. Recommendations and performance impact scores are intentionally conservative to ensure that maximum system performance is maintained.

Also, all utilizations and performance impact scores are derived from the four-hour windows. When a single value is returned, the value return is the value of the most performance-limiting four-hour window value for the most limiting component on the storage system. For example, during new provisioning, the impact of adding a workload to a front-end port could take the utilization of that port from ~30% utilized to ~35% for most time windows, but 32% to 37% on Tuesdays between 8AM and 12PM UTC. In this case, the new port utilization would be reported as 37% because that is the most limiting four-hour window value.

The four-hour window implementation allows Workload Planner to pack workloads more efficiently and realistically onto storage systems regarding inverse performance fluctuations. For example, a workload that is heavy on weekends, but light during weekdays performs better with a second workload that is light on weekends and heavy on weekdays than a workload that is distributed similarly to itself.

One drawback to the four-hour window implementation is that it is sensitive to peaks in data. If unexpected performance impact, system headroom, and/or service level compliance results are observed, see Manage System Data Exclusion Windows for help on remediating these issues.

The following workload monitoring, planning, and provisioning tasks can be performed from Unisphere for PowerMax: