Pod rightsizing workload optimization

Prev Next

Pod rightsizing workload optimization

This topic describes optimizing a workload with Pod rightsizing.

A workload is optimized after the following steps:

  1. You apply Pod rightsizing to a workload together with an optimization strategy.
    Each strategy calculates recommende CPU and memory requests using usage percentiles from the last 24 hours plus a buffer to ensure performance during demand fluctuations.

  2. Pod Rightsizing changes the allocated resources according to the selected optimization strategy.

This process completes within an hour of applying the solution.

You can choose from the following resource optimization strategies:

Strategy

CPU and RAM allocation

Criteria

Savings benefit

Relevant for this kind of workload

Default

Reduced to the lowest safe level

  • CPU
    P95 + 20% buffer

  • Memory
    P99 + 30% buffer

Highest savings

Most workloads

Conservative

Reduced moderately

  • CPU
    Max + 30% buffer

  • Memory
    Max + 30% buffer

Moderate savings

Sensitive or stateful workloads

The following table demonstrates Default optimization strategy recommendations:

Over-provisioned

Under-provisioned

Current
CPU request
(last 24 hours)

2.0 cores

1.8 cores

Actual
P95 CPU usage

(last 24 hours)

0.6 cores

1.0 cores

Strategy buffer calculation (20%)

0.6 x 0.2 = 0.12 cores

1.8 x 0.2 = 0.36 cores

Recommendation

0.6 + 0.12 = .72 cores

1.8 + 0.36 = 2.16 cores

Result

Reduce from 2.0 to 0.72 cores, saving 64% with safety buffer for spikes

Increase from 1.0 to 2.16 cores, preventing throttling and performance issues

A workload could have both over-provisioned and under-provisioned recommendations, one for RAM and the other for CPU, or vice-versa.

After the initial resource reduction, Kompass continues to analyze and adjust the resources to ensure that managed workloads are being optimized according to the selected strategy. Changes may be made every hour or when an optimization strategy is changed.

To apply Pod rightsizing to a workload, see Apply and manage Pod rightsizing.

After Pod rightsizing is active on a workload, you can see the impact on the workload in terms of CPU, RAM, cost, and other metrics. You can also change the selected optimization strategy or remove the solution entirely. For more info, see Manage Pod rightsizing on a workload.

Changes made by the solution are logged in the Audit log.

For information about supported environments and limitations, see Compute optimization.

Where to go from here