Resource Analysis, Actions, and Optimization
Each cycle, the market compares demand from workloads against supply from hosts, datastores, and clouds, then emits actions when a buyer can be served better elsewhere or in a smaller shape.
How analysis works
- Commodities include CPU, memory, storage, IOPS, latency, and network. Prices rise as a seller fills, so an overloaded host becomes expensive and buyers move away.
- Turbonomic works from a statistical peak over the observation window rather than instantaneous samples, which is why a brief spike rarely triggers a resize while a sustained peak does.
- Action families: resize up or down, move, provision and decommission (including cluster scale and node provisioning), suspend and start, plus cloud purchasing actions such as reserved instance and savings plan recommendations.
Reading actions
- Each action shows current versus projected utilization, a risk assessment, and for cloud an estimated saving.
- Pending Actions are not yet executed; Action History records executed and failed actions with the orchestrator result.
- Plans let you simulate: add hosts, move workloads, change instance types, and compare utilization and cost before touching policy.
Worked example
A host sits at 90 percent peak memory, yet no move is recommended. Reasoning: every other host in the cluster is at a similar level, so there is no cheaper seller; the market's answer is to resize the largest VMs down or add capacity, which surfaces as resize or provision actions instead of moves. A placement policy will not fix it.
Traps
Reading average utilization as the driver; treating estimated savings as guaranteed; assuming resizes need no maintenance window; confusing VM actions with container actions such as resize container and scale workload.