Payments are experiencing issues due to temporary restrictions in Russia. If your payment does not go through, please submit a support request.Our support team is available 24/7 — we are always here to help with hosting and server issues.We are now accepting requests for dedicated server rental and colocation services in our data center.Reminder: we recommend enabling backups for additional data protection.A new VPS/VDS lineup with NVMe storage and improved performance is now available.Maintenance work on some servers has been completed. All services are operating normally.
Article3 min readViews0

What Load Average Means and When a High Value Requires Investigation

We explain Linux load average: why it cannot be read as CPU percentages and how to correlate it with the number of processors and resource wait times.

Server internal components: CPU cooling and storage drives
In this article

The server report shows load average values, and the highest one looks alarming. Without context, this tells us little. Load depends on the number of tasks that are running, waiting for the CPU, or in uninterruptible sleep, not just on computations.

Read the three intervals

On Linux with procps, run:

uptime

At the end of the line, load averages for one, five, and fifteen minutes are shown. These are smoothed values, so they do not have to drop instantly after a heavy task finishes. Comparing short and long intervals helps see the direction of change, but does not name the cause.

If the one-minute value is higher than the others, the load recently increased. If the long-term value remains high while the short-term value drops, the system may be exiting a peak. To resolve the issue, you still need current processes and signs of resource waiting.

Account for available CPU

In a system with GNU Coreutils, you can view the number of computational units currently available to the process as follows:

nproc

This is a useful reference, but not a universal divisor for all environments. Containers, CPU limits, and virtual machine placement alter the picture of available resources. Do not declare a server healthy simply because the average load is lower than the number of logical processors.

Hypothetical example: the same load value on a small virtual machine and on a multiprocessor node implies different conditions for CPU contention. Even a powerful server may respond slowly if processes are waiting for a single overloaded disk.

Identify the nature of the wait

Supplement the picture with a brief observation:

vmstat 1 6

The column r reflects tasks ready to run, while b shows processes in uninterruptible sleep. For current activity, examine the subsequent rows, taking into account the specifics of the first report. Correlate this with CPU utilization and application latency.

A large number of waiting processes does not automatically indicate a physical disk failure. The source of the wait may be various resources or environmental factors. This result only suggests which hypothesis to test next.

Do not set a single threshold for all servers

The threshold must account for the typical load profile and its impact on the service. A brief spike during a scheduled background task differs from a persistent queue where checkout takes too long. It is useful to monitor response time, errors, and dependency availability simultaneously.

For comparison, record the current task, the number of available processors, all three load values, and the observation period. A single screenshot without a date or context is difficult to correlate with import logs and database monitoring.

When diagnostics yield a result

You must be able to explain whether the CPU queue is growing or the system is waiting for another resource, which workflow coincided with the spike, and how users noticed it. After fixing the issue, repeat the observation under comparable load. A reduction in the panel value alone does not prove that the cause of the slow site has been eliminated.

When evaluating a virtual machine, specify the CPU limit allocated to it. The number of visible processors and the guaranteed compute share may differ.

Discussion 0

Share your experience and ask questions. Comments without links appear after editorial review.

No comments yet. Start the discussion.