Is My Instance Being Affected by Host Over-provisioning?

These panels let you decide, on your own, whether the underlying host is honest about the resources you paid for. If they are clean, the problem is inside the guest and you should move on to Q2.

Hypervisor CPU Queue Time

The time a vCPU was ready to run but was not actually scheduled onto a physical core, measured by the hypervisor. This is the authoritative measurement of CPU steal — reported from the host side, where it cannot be hidden.

If this panel is non-trivial, the host is over-committed: another tenant or the host itself is competing for the same physical cores. Open a support ticket and reference this panel.

CPU Other

The same phenomenon, observed from inside the guest. The "Other" panel sums the less common CPU states reported by /proc/stat, including steal time as the guest sees it.

This panel and Hypervisor CPU Queue Time should always agree in shape. We expose both deliberately so you can cross-check what your existing in-guest tooling (top, sar, node_exporter, etc.) is telling you against the host's view.

CPU Pressure / Memory Pressure / I/O Pressure

Pressure Stall Information ("PSI") is a Linux kernel feature that reports the share of time tasks were stalled waiting for a resource. ZEC exposes three panels, one per resource:

• CPU Pressure — vCPUs were runnable but waiting on a physical core.

• Memory Pressure — the workload was stalled in memory reclaim or allocation.

• I/O Pressure — the workload was stalled waiting on the block backend.

Each panel has two lines:

• some — at least one task was stalled.

• full — every task was stalled (i.e., the workload made no forward progress at all). full does not apply to CPU; only memory and I/O expose it.

Pressure is the clearest single signal for "the host is the problem." Hypervisor CPU Queue Time tells you how much CPU was stolen; Pressure tells you how much that hurt. A short spike of stolen CPU on an idle workload is harmless; a small amount of stolen CPU on a workload already saturating its cores can drive the full line to 100% and stall the application completely.

When making an over-provisioning claim, screenshot both Hypervisor CPU Queue Time and the relevant Pressure panel. Together they are enough evidence for a support ticket to skip the back-and-forth.

API Reference

All instance metrics use DescribeInstanceMonitorData with the metricType values below.

Console Panel	metricType	Notes
Hypervisor CPU Queue Time	HYPERVISOR_QUEUE_TIME
CPU Other	PROC_CPU_OTHER
CPU Pressure	—	Console only
Memory Pressure	—	Console only
I/O Pressure	—	Console only