VMware ESXi and vSphere Cluster Management
Real-Time and Historical Performance Statistics in VMware vSphere
Learn how vSphere collects, stores, rolls up, retains, and displays real-time and historical performance statistics for ESXi hosts and virtual machines.
VMware vSphere performance statistics are time-based measurements collected from ESXi hosts and virtual machines. They help administrators monitor current behavior, troubleshoot resource problems, analyze workload patterns, and plan future capacity.
The most important distinction is between real-time statistics, which preserve recent detail for immediate investigation, and historical statistics, which retain summarized data for longer-term analysis.
What vSphere Performance Statistics Measure
A performance statistic is a measurement of a specific aspect of system behavior. That measurement is often called a performance counter. Examples include CPU usage, memory usage, network throughput, and datastore or virtual disk latency.
vSphere records these measurements over time for ESXi hosts and virtual machines. A single reading can show what is happening at one moment, while a sequence of readings reveals workload changes, sustained utilization, recurring peaks, and possible contention.
- Monitoring: Observe resource use and system health.
- Troubleshooting: Investigate a slowdown or resource problem occurring now or recently.
- Trend analysis: Identify recurring behavior and workload growth.
- Capacity analysis: Determine whether compute, storage, or network resources will meet future demand.
Use high-resolution data for immediate troubleshooting and longer-range data for patterns and planning. A long-term chart is not a replacement for detailed evidence collected during a short incident.
Real-Time Statistics
Real-time statistics are recent performance measurements collected at 20-second intervals. ESXi hosts collect these measurements for the host itself and for its virtual machines.
The real-time view covers approximately the most recent hour. At one measurement every 20 seconds, one hour contains approximately:
60 minutes × 60 seconds ÷ 20 seconds = 180 data pointsReal-time information is available from local flat-file storage on the ESXi host and in vCenter Server memory. This arrangement supports quick access to recent measurements without requiring every near-term point to be retained indefinitely in the vCenter Server database.
Real-time charts are particularly useful when a performance issue is occurring now or occurred very recently. For example, a short CPU spike, a burst of network traffic, or a brief storage-latency event may be visible in the 20-second points even when it disappears into a broader historical aggregate.
Example: Investigating a Current Virtual Machine Slowdown
- Select the affected virtual machine in the vSphere Client.
- Open its performance monitoring view and choose the real-time range.
- Review CPU, memory, disk, and network charts around the reported incident time.
- Look for brief spikes, sustained pressure, or changes that align with the user-visible slowdown.
- Correlate the chart with workload activity, configuration changes, backups, migrations, or infrastructure events.
Historical Statistics
Historical statistics are performance measurements retained for longer periods and represented by progressively broader sampling intervals. vCenter Server stores historical statistics in its database.
The standard historical views are the past day, past week, past month, and past year. As the selected period becomes longer, vSphere uses larger intervals and rolled-up values so that the database and charts remain manageable.
An older chart point is therefore an aggregate for its interval, not necessarily one instantaneous reading. The exact meaning of an aggregate depends on the counter and the statistic being displayed, but the key principle is that historical points summarize a period of time.
Sampling Intervals and Retention
A sampling interval is the time between displayed or stored data points. A retention period is how long data remains available at a particular interval. The selected time range determines the level of detail that the chart can show.
For example, a past-day chart with five-minute points contains:
24 hours × 60 minutes ÷ 5 minutes = 288 pointsA past-year chart with one point per day contains approximately 365 points. These figures contrast with real-time data, where 20-second measurements produce approximately 180 points in one hour.
How Statistics Are Rolled Up
A rollup is the aggregation of finer-grained samples into a broader time interval. The general progression is:
- ESXi collects near-term measurements at 20-second intervals.
- Recent measurements are available in the real-time view.
- Older measurements are summarized into five-minute, 30-minute, two-hour, and daily historical intervals.
- vCenter Server stores the retained historical rollups in its database.
Rollups reduce database growth and processing requirements while preserving useful information for longer-term analysis. Without aggregation, retaining every 20-second point for every host, virtual machine, and counter would require substantially more storage and processing.
Rollups also affect interpretation. A brief resource spike may be obvious in a real-time chart but may be averaged or otherwise obscured in a two-hour or daily point. Conversely, a sustained increase is more likely to remain visible in a long-range chart.
Real-Time Versus Historical Statistics
Statistics Levels and Collection Configuration
vCenter Server can be configured to control how much performance information is collected and retained for historical reporting. Statistics levels control the breadth and detail of performance counters available in historical data.
Higher-detail collection can make more counters available and improve diagnosis, reporting, and capacity analysis. The trade-off is increased vCenter Server database storage, processing, and maintenance requirements. The appropriate setting depends on operational needs rather than a universal preference for maximum detail.
Reviewing Collection Settings
- Navigate to the vCenter Server performance or statistics configuration settings in the vSphere Client.
- Review the collection level assigned to each historical interval.
- Review the sampling intervals and retention periods.
- Compare the configured policy with monitoring, reporting, troubleshooting, and capacity-management requirements.
- Before increasing collection detail, assess database capacity and processing requirements.
- After a change, verify that the policy provides the required metric detail for the required period.
Viewing Statistics in the vSphere Client
The general monitoring workflow is similar for an ESXi host and a virtual machine:
- Select the ESXi host or virtual machine in the vCenter inventory.
- Open the performance monitoring view.
- Choose a time range: real-time, day, week, month, or year.
- Select a metric category such as CPU, memory, disk, or network.
- Choose the relevant performance counters.
- Compare charts across time ranges to separate a current event from a recurring or long-term issue.
Always interpret the chart using its interval. A real-time chart can help establish when a short event occurred. A past-day chart can show whether activity repeats throughout the day. A past-month or past-year chart can show sustained growth, but it cannot reliably prove the exact maximum of a brief event.
Choosing a View for the Monitoring Goal
Using Statistics for Troubleshooting and Planning
Current CPU, Memory, Storage, or Network Problems
Use real-time charts to investigate CPU contention, memory pressure, storage latency, or unusual network traffic. Examine the timing and duration of changes rather than relying on one point. Then correlate the measurement with virtual machine activity, host operations, configuration changes, and infrastructure events.
Recurring Workload Patterns
Use the past-day view to identify daily peaks and compare their timing with backups, batch jobs, replication, or user activity. Use the past-week view to determine whether the pattern repeats on particular days or during scheduled operations.
Capacity Growth
Use past-month and past-year views to evaluate sustained CPU, memory, disk, and network utilization. Coarser rollups are appropriate for identifying growth and long-term demand, but do not treat a daily point as proof of a brief maximum spike.
For reliable conclusions, compare multiple counters and multiple time ranges. A rise in CPU usage might reflect legitimate workload growth, while increased storage latency or memory pressure could indicate a different bottleneck.
Troubleshooting Common Statistics Questions
A Short Problem Does Not Appear in a Long-Range Chart
Historical rollups use broad intervals. A short spike may be averaged or aggregated into a less obvious value. Use real-time data or the shortest historical range that is still retained, and collect evidence close to the time of the incident.
A Detailed Historical Counter Is Missing
Check whether the selected range is historical rather than real-time. Then review the configured statistics level and retention settings. The counter may not have been collected at the required detail, or its data may have aged beyond the retention period.
The Chart Is Less Granular Than Expected
The selected range probably uses historical rollups instead of 20-second real-time samples. Select a shorter range if that data is still available, and identify the interval associated with the view before interpreting its values.
Historical Reporting Is Increasing Database Concerns
Higher collection detail and longer retention increase the volume of historical statistics stored by vCenter Server. Review which counters and periods are required for operations and reporting, then assess database capacity before increasing collection levels.
Key Takeaways
- Real-time statistics use 20-second measurements and cover approximately the latest hour.
- Historical statistics are stored in the vCenter Server database and are retained as progressively broader rollups.
- The standard historical intervals are five minutes for the past day, 30 minutes for the past week, two hours for the past month, and one day for the past year.
- Real-time data is best for brief, current events; historical data is best for recurring patterns, growth, and capacity planning.
- A rolled-up point represents an aggregate for an interval, not necessarily an instantaneous reading.
- Statistics levels and retention settings affect which historical counters are available and how much database capacity is required.
- Always correlate charts with workload activity, configuration changes, and infrastructure events.
Related Learning
For broader platform context, see the VMware ESXi online course. For inventory and monitoring prerequisites, review adding an ESXi host to vCenter Server inventory and communication between vCenter Server and ESXi. Performance-based operational responses also connect with alarm trigger types and acknowledging an alarm.