Steal time (st in top) is the percentage of time your virtual CPU was ready to run but the hypervisor was serving other guests. Sustained steal above about 10% means the physical host is oversubscribed or your instance has used up its burst credits. You cannot fix it inside the VM: resize, switch to dedicated CPU, or ask the provider to move you.
Symptoms
- top shows st above 5–10%
- Performance drops while your own CPU usage looks unchanged
- Slowness at the same times each day
How to diagnose it
Measure steal
Check the st value in top and the st column in vmstat.
vmstat 1 10Check per-core steal
%steal per CPU.
mpstat -P ALL 1 5Check burst credits on burstable instances
On AWS T-series, steal appears when CPU credits run out. Check CPUCreditBalance in CloudWatch.
Common causes and fixes
Noisy neighbours on an oversubscribed host
Ask the provider to migrate the VM, or move to a dedicated-CPU plan.
Burstable instance out of credits
Move to a non-burstable instance type or enable unlimited mode if the cost is acceptable.
VM too small for the workload
More vCPUs give the scheduler more room and reduce the impact of steal.
Prevent it happening again
- Alert when CPU_steal is above 10% for 15 minutes.
- Keep steal history as evidence when raising a ticket with your provider.
Set it up: AWS EC2 monitoring · DigitalOcean monitoring · Hetzner monitoring.
FAQ
What is a normal steal time?
Close to 0%. Occasional small spikes are normal on shared hosts; sustained values above 10% hurt performance.
Does steal time show on dedicated servers?
No. It only exists on virtual machines.
Related guides
Get alerted before this becomes an outage.
Install the agent in one line. Statance watches CPU, memory, disk and services, and its AI assistant helps find the cause.