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

Disk is lagging: how to read iostat without universal thresholds

We examine latency, queue depth, and I/O operations, accounting for SSDs, virtual disks, and the difference between the first report and subsequent intervals.

Solid-state drive contacts and components
In this article

The CPU is not fully utilized, yet the site responds slowly, and background tasks take longer than usual. A likely cause is storage wait time. iostat helps reveal device activity, but a single high percentage does not explain the issue. Multiple intervals and correlation with the workload are needed.

Take a short series

This guide assumes Linux with the sysstat package installed. The viewing command does not change device settings:

iostat -xz 1 6

The extended output parameter adds device metrics, while excluding inactive rows reduces noise. The first report typically contains cumulative values since boot; subsequent reports cover observation intervals. Therefore, for a current incident, do not limit yourself to the first table.

Verify device names against the mount scheme. If an application runs on a logical or network volume, monitoring a random physical disk may not answer the right question.

Separate throughput from latency

Transfer speed indicates how much data is processed, while the operation count shows how many requests are handled. The same megabyte throughput can consist of large sequential reads or numerous small operations. For an application, these scenarios carry different costs.

Wait metrics, such as r_await and w_await, reflect the average latency of the corresponding requests, including queue wait time and service time. Column names depend on the version. The average can mask rare long operations, so compare it with application latency and more detailed monitoring if available.

Be cautious with utilization percentage

%util cannot be considered a universal measure of the limit for any modern storage. Devices that handle requests in parallel differ from a single sequential mechanical disk. The value alone does not indicate how much useful throughput remains.

On a virtual server, provider limits and overall storage levels impose additional constraints. Low visible data throughput does not rule out operation count limits or latency occurring outside the guest system.

Align with tasks

Example: latency spikes coincide with backup runs and mass imports. It is useful to check whether they compete for the same resource. Running a speed test over this load adds new operations and complicates conclusions, so initial diagnostics should remain observational.

If activity is high but user scenarios have not degraded, this may be normal productive work. If site latency rises with moderate throughput, examine the operation characteristics and additional constraints. Do not declare a device faulty based on a single table.

Results and re-verification

Record the time interval, device, load type, and metrics from several consecutive rows. Note parallel background jobs. After the selected change, repeat the observation under comparable conditions and verify user response time.

If you need to contact a provider, it is more useful to provide the exact interval and measurements than to state 'the disk is slow.' Diagnostics should narrow the scope of the problem: a queue within the guest system, competing tasks, or constraints of the external storage. The result is evaluated based on stable application performance, not by achieving an arbitrary percentage.

Discussion 0

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

No comments yet. Start the discussion.