Forecasting Nutanix Capacity with ClusterCollie AI

Abstract

ClusterCollie AI is a capacity forecasting dashboard for Nutanix infrastructure. It ingests historical utilization telemetry from the Nutanix Prism Central API and applies statistical and machine-learning trend models to project compute, storage, and network headroom forward across a configurable horizon. This paper describes the problem the tool addresses, the forecasting methodology, system architecture, data handling practices, and known limitations of the current beta release.

1. Introduction

Capacity planning for hyperconverged infrastructure is a continuous, moving target. A Nutanix cluster's compute, storage, and network resources rarely saturate at the same rate, and administrators are often forced to reconstruct a capacity trend by hand from periodic Prism Central snapshots or exported spreadsheets. ClusterCollie AI automates this by treating capacity planning as a time-series forecasting problem: given a history of utilization samples, project forward and flag when a resource is likely to cross a defined threshold.

2. Problem Statement

Three recurring failure modes motivated this project:

Each of these is, at its core, a forecasting gap: existing tooling reports current state well but does not project it forward with enough lead time to act.

3. Data Sources

ClusterCollie AI's forecasts are derived from historical utilization metrics pulled directly from Nutanix Prism Central, either via its native API or through the Nutanix MCP server. Metrics currently in scope include:

• Host-level CPU and memory utilization
• Storage container capacity, growth rate, deduplication, and compression ratios
• Network throughput and bandwidth utilization patterns

For evaluation without live cluster access, the application also ships with a sample dataset that exercises the same forecasting pipeline.

4. Forecasting Methodology

Each tracked metric is treated as an independent time series. The forecasting pipeline:

5. System Architecture

The application is built as a local-first Streamlit dashboard. At a high level:

• Connection layer: authenticates to Prism Central (or the Nutanix MCP server) using read-only credentials supplied via local environment configuration.
• Ingestion layer: retrieves and caches historical utilization data for the selected clusters.
• Forecasting engine: applies the trend models described in Section 4 to produce per-metric projections.
• Presentation layer: renders forecasts, confidence bands, and threshold warnings in an interactive browser-based dashboard.

By default, cluster telemetry and credentials are processed locally and are not transmitted to Project Research LLC infrastructure.

6. Limitations

7. Governance & Responsible Use

ClusterCollie AI's forecasting outputs are governed under Project Research LLC's published AI Governance Policy, which is aligned with the structure of the NIST AI Risk Management Framework. Human review by a qualified infrastructure planner is required before forecast output informs procurement, budgeting, or workload migration decisions.

8. Roadmap

Planned areas of continued development include expanded metric coverage, refinement of trend models based on beta usage data, and improved anomaly detection for sudden utilization shifts. Feedback from beta users directly informs prioritization — see the Contact page to report issues or suggest improvements.

9. Conclusion

ClusterCollie AI reframes Nutanix capacity planning as a forward-looking forecasting problem rather than a point-in-time reporting exercise. By combining Prism Central telemetry with statistical and machine-learning trend models, it aims to give infrastructure teams enough lead time to act on capacity risk before it becomes an incident.

Authors & Contact

Published by Project Research LLC. Questions about this white paper or the forecasting methodology it describes can be directed to admin@projectresearchllc.com.