NodeFit: node right-sizing

NodeFit is an advanced feature that provides a granular view of resource allocation across the various nodes supporting your Kubernetes clusters

NodeFit overview

NodeFit is an advanced feature that provides comprehensive visibility of the entire Kubernetes environment at the infrastructure level. It helps in understanding the behavior of specific node groups and node types, and provides actionable data-driven insights, optimizing the underlying infrastructure of the workloads. Infrafit offers a clear historical view of node utilization across the entire environment, enhancing the optimization process.

This level of visibility allows you to identify areas with idle space and gain insights into adjusting node sizes based on actual resource utilization and selecting the optimal node types to support the needs of your workloads. Additionally, NodeFit provides a unique view into scheduling results, showing what workloads are being scheduled to each node group, what is preventing scaling down, and what causes overcommitment over time.

NodeFit & PodFit connectivity

NodeFit and PodFit together offer a comprehensive solution that provides a multidimensional approach to reduce wasted resources in your nodes. When drilling down into individual node types or node groups, NodeFit provides a unique view into scheduling results, showing which workloads are being scheduled to the node group, what is preventing scaling down, and what causes overcommitting over time.

While PodFit helps you right-size your workloads based on their actual resource utilization, NodeFit provides data-driven recommendations that help you choose the optimal node type to best serve the needs of your workloads.

This provides insights that can help you more efficiently bin-pack and size your node to maximize resource utilization. It also helps you lower your overall node count while improving the efficiency of your node-autoscalers (like Karpenter or Cluster Autoscaler) to keep your costs optimized. Learn more about Karpenter optimization.

By combining the capabilities of both features, you gain a comprehensive understanding of your infrastructure and workloads, enabling more efficient management and cost reduction.

Key benefits

• Integrated Insights: NodeFit provides detailed visibility into node groups and node types, while PodFit offers insights into workload scheduling and resource allocation. Together, they give you a holistic view of your environment.

• Optimized Resource Allocation: By analyzing the data from NodeFit and PodFit, you can make informed decisions on adjusting node sizes, selecting the best nodes for your workloads, improving bin-packing, and optimizing the performance of node-autoscalers (e.g., Karpenter or Cluster Autoscaler).

• Enhanced Efficiency: Identify workloads that prevent scaling down or cause overcommitment and adjust resource allocations accordingly. This helps in lowering the overall node count while maximizing resource utilization.

• Cost Reduction: With a clear understanding of resource usage and waste, you can implement strategies to reduce costs and improve the efficiency of your infrastructure.

• Maximize Karpenter: Get the most value from Karpenter by combining workload-level insights with node-level visibility. By right-sizing workloads and analyzing scheduling patterns, node utilization, and NodePool configurations, PerfectScale identifies opportunities and provides actionable recommendations to improve bin-packing, expand instance flexibility, reduce unused capacity, and enhance consolidation efficiency. Learn more about Karpenter optimization.

How it works

• Workload Scheduling Visibility: NodeFit provides a unique view into scheduling results, showing which workloads are being scheduled to each node group, what prevents scaling down, and what causes overcommitment over time.

• Resource Utilization Insights: PodFit complements this by offering detailed insights into workload runtime and resource usage, helping you identify areas of improvement.

• Data-Driven Recommendations: Together, NodeFit and PodFit provide actionable recommendations to optimize your Kubernetes environment, ensuring that resources are allocated efficiently and costs are minimized.

By leveraging the combined power of NodeFit and PodFit, you can achieve a more efficient, cost-effective, and well-optimized Kubernetes infrastructure. We will explore how to leverage these capabilities throughout the documentation.

Upper Panel

NodeFit upper panel

Tenant name

Displays the name of the account and enables you to switch between different accounts quickly.

Cluster name

Display the name of the selected cluster.

Cluster autoscaler

Displays the autoscaling solution, such as Karpenter or Cluster Autoscaler, detected in the selected cluster.

Optimization Policy

Displays the cluster's optimization policy. Optimization policy allows you to specify how your resources should be allocated in order to support the individual needs of your workloads. Define the policies that best suit your environment and business goals, depending on whether you want to maximize cost savings or provide extra headroom to maintain the resilience of mission-critical services.

• MaxSavings - maximum cost savings, the best for non-production environments

• Balanced (default) - optimally balances cost and resiliency

• ExtraHeadroom - the best fit for latency-sensitive environments

• MaxHeadroom - keeps the environment above the highest spikes

The Optimization Policy can be set for the entire cluster and a specific workload. The workload's Optimization Policy takes precedence and will override the value defined at the cluster level.

If a custom policy is set through the exporter when installing the PerfectScale Agent, it cannot be modified in the UI afterward. You can still change the policy by upgrading the exporter with the new value, or you can return it to the default by upgrading the exporter without specifying any value (this will also enable the option to change the custom time window through the UI).

Discover more about customizing the Optimization policy here.

Timeframe

Allows you to check the data for a specific time period: click on the drop-down list in the upper right corner and select one of the options.

Export

This feature allows you to seamlessly analyze and effortlessly share your data by exporting it into a .csv file. Click the Export button, and the data will be exported to your local machine in a few seconds.

Nodes resource utilization

PerfectScale provides comprehensive GPU visibility when GPU nodes are detected in the cluster. This visibility enables you to monitor GPU usage in real time, identify underutilized or idle resources, and make informed decisions to optimize GPU allocation and reduce waste. Learn more about GPU optimization here.

Nodes utilization

Hover over the specific time point on the chart to view data for that time

Node Groups/Node Types selector

This selector allows easy switching between node groups and node types, giving you different perspectives of your infrastructure for more comprehensive analysis.

Utilization

A drop-down selector allows you to choose the resource utilization percentile (affects graphs as well as the Utilization chart in the table below), helping you better understand the resource utilization patterns.

Cost per Node Group/Node Type

Displays the cost of the Node Groups or Node Types over the selected timeframe.

The diagram displays the top 10 node groups or node types with the highest costs

CPU over time

Displays the allocated, requested, and used amount of CPU (cores) with the selected utilization percentile over the selected timeframe.

Memory over time

Displays the allocated, requested, and used amount of Memory (GB) with the selected utilization percentile over the selected timeframe.

GPU and GPU Memory Over Time

When PerfectScale detects active GPU resources within a cluster, it automatically enables GPU and GPU memory widgets and utilization insights in the UI. Learn more about GPU resources optimization here.