Add Autonomous Optimization of Compute & Storage Resources as New Pattern

[johnjamie]

Describe the pattern you'd like to propose
Autonomous Optimization of CPU, memory, storage reduces the hardware needs and emissions for a given workload. It uses machine learning to uncover the most efficient configuration for resources and avoids typical overprovisioning outcomes that occur when manual parameters are used. Specifically, application behavior at different traffic and resource levels is analyzed to determine the most efficient configuration. Performance/availability constraints are used to ensure business needs are met while reducing resource utilization. Autonomous optimization can be configured to optimize against resource cost, resource usage and/or direct emission estimates.

This pattern enables the implementation of other patterns including:

• Scale Kubernetes workloads based on relevant demand metrics (https://patterns.greensoftware.foundation/catalog/cloud/scale-kubernetes-workloads-based-on-events). Autonomous optimization users demand metrics as an input to determine how and when to scale, including predictive capabilities that can reduce the amount of resource buffers.
• Optimize average CPU utilization (https://patterns.greensoftware.foundation/catalog/cloud/optimize-avg-cpu-utilization). Autonomous optimization evaluates the utilization and optimizes for maximum efficiency.
• Match utilization requirements of virtual machines (VMs) (https://patterns.greensoftware.foundation/catalog/cloud/match-utilization-requirements-of-vm). Autonomous optimization evaluates the utilization and selects the most efficient resource type.
• Scale down Kubernetes applications when not in use(https://patterns.greensoftware.foundation/catalog/cloud/scale-down-kubernetes-workloads). Autonomous optimization provides more efficient inputs to autoscaling tools than can be achieved by manual methods.
• Optimize storage utilization (https://patterns.greensoftware.foundation/catalog/cloud/optimise-storage-resource-utilisation). Autonomous optimization adds capacity only when needed improving storage utilization rates. It can also subdivide large volumes to smaller volumes which allow performance goals to be met with less storage capacity.

Describe specific emission impact from this pattern
SCI Impact - SCI = (E * I) + M per R

Autonomous optimization impacts SCI as follows:
E: The goal is to "reduce demand", such as for k8s the number of pods running during or out of office hours, which reduces overall consumed power by the application. In addition, optimizing average CPU & storage utilization can reduce the amount of energy needed to support the systems traffic from the reduced CPU & storage requirements.
M: Optimized metrics such as average CPU & storage utilization can reduce the amount of resources needed which will decrease the amount of embodied carbon required to support the system.

References to this pattern
Gartner coverage of real-time optimization; prediction of growth in "Predicts 2023: Observing and Optimizing the Adaptive Organization", 17 November 2022 - link (subscription required). Gartner notes "
By 2026, organizations performing real-time cost or performance optimization of cloud-based workloads will rise from less than 20% in 2022 to 50%."
One day conference on autonomous (here)
Kubernetes workshop with focus on cost as the optimization goal (one hour, here)

Considerations
This pattern does not require application code changes.
This pattern can be achieved by developing internal company tools or using commercial services. A range of tools with varying autonomous capabilities are listed on the CNCF roadmap under "Continuous Optimization" here)

Additional context
Example reduction in CPU usage (42%) and memory (38%) for optimization at the workload level for Kubernetes:

Disclosure
I work for Sedai, a provider of autonomous optimization.