Performance Baseline

This domain captures short-horizon operational signals that help teams understand the current state of the environment.

It is a baseline, not a full long-term observability platform.

What Ranger Collects

The performance-baseline domain should document:

• host CPU and memory utilization snapshots
• storage IOPS, throughput, latency, cache, and health-adjacent signals
• network throughput and error indicators
• active alert or Health Service summaries
• recent event patterns that materially affect interpretation of the current state

Manifest Sub-Domains

The v1 collector writes to these named sections of the performance manifest domain:

Sub-domain	Content
nodes	Per-node CPU utilization, memory pressure, and uptime snapshot
compute	Cluster-wide vCPU and memory allocation vs. capacity with overcommit indicators
storage	Storage IOPS, throughput, latency, and cache hit-ratio baselines
networking	Network throughput and adapter error counts
outliers	Nodes or workloads that significantly exceed utilization norms
events	Recent Health Service events and cluster health faults
summary	Aggregate performance risk signals — high-CPU nodes, high-memory nodes, and alert counts

Current Collector Depth

Current v1 collection also covers:

• RDMA and host-network performance counters where available.
• CSV cache and storage-latency context used for platform health interpretation.
• Event-log aggregation and outlier detection across the nodes.
• Point-in-time performance baselines that feed management and technical report sections.

Why It Matters

Ranger should not stop at static inventory. A short-horizon operational baseline helps explain whether a healthy-looking design is currently under stress, degraded, or simply under-observed.

Connectivity and Credentials

Requirement	Purpose
WinRM / PowerShell remoting	Host-side performance and event data
Cluster credential	Required
Optional Azure credential	Useful for Azure Monitor, metrics, alerts, and HCI Insights correlation

Default Behavior

This domain should run by default when cluster credentials are available.

Azure-side observability overlays should be partial or skipped when Azure credentials are missing.

Variant Behavior

Hyperconverged

Baseline host, storage, and network metrics apply directly.

Disconnected Operations

Monitoring paths differ because local control-plane observability matters more than public Azure telemetry assumptions.

Multi-Rack Preview

Performance interpretation may need to respect managed networking and SAN-backed storage rather than assuming standard hyperconverged behaviors.

Example Manifest Data

A successful collect produces entries like this:

{
  "id": "managementPerformance",
  "status": "success",
  "domains": {
    "performance": {
      "nodes": [
        { "host": "tplabs-01-n01", "cpuUsagePercent": 12.4, "memoryUsedGB": 148,
          "memoryTotalGB": 256, "uptimeDays": 14 },
        { "host": "tplabs-01-n02", "cpuUsagePercent": 8.1, "memoryUsedGB": 132,
          "memoryTotalGB": 256, "uptimeDays": 14 }
      ],
      "storage": {
        "iopsRead": 4821, "iopsWrite": 2310,
        "latencyReadMs": 0.4, "latencyWriteMs": 0.6,
        "cacheHitRatePercent": 94.2
      },
      "outliers": [],
      "summary": { "highCpuNodeCount": 0, "highMemoryNodeCount": 0, "activeAlertCount": 0 }
    }
  }
}

Common Findings

Finding	Severity	What it means
Node CPU utilization above 80%	Warning	Host is under significant compute load at time of collection; investigate workload distribution
Node memory utilization above 90%	Warning	Memory pressure detected; workload consolidation may be needed
Storage latency above threshold	Warning	Read or write latency is elevated; investigate cache hit rate and disk health
Storage cache hit rate below 80%	Warning	Cache is not effective; working set may exceed cache capacity
Active Health Service alerts	Warning	Cluster has open health faults; investigate before treating run data as a clean baseline

Partial Status

status: partial on the performance collector means:

• Per-node CPU/memory snapshots succeeded on some nodes but not others
• Azure Monitor overlay (HCI Insights, DCR state) failed while local performance counters succeeded — local baseline is still valid
• Event-log aggregation timed out while compute and storage metrics succeeded

Local performance snapshots are collected independently of Azure Monitor. Azure Monitor absence does not invalidate the on-premises performance baseline.

Domain Dependencies

Depends on the cluster-and-node domain for a node list. Azure Monitor overlay sections also depend on valid targets.azure configuration and an active Azure credential.

Evidence Boundaries

• Direct discovery: host-side utilization and event snapshots
• Azure-side discovery: Azure Monitor, alerts, DCRs, and HCI Insights context
• Manual/imported evidence: operator-supplied baselines or SLO context when needed

v1 and Future Boundaries

v1 should provide a useful snapshot and anomaly framing.

It should not promise long-term trending, forecasting, or full APM-style workload performance analysis.