30-Second Fix: Check the %RDY metric
Open vCenter, find your VM, go to the Performance tab, select Advanced, and add the metric "CPU Ready" as a chart line. If it's consistently above 5-10% of a single vCPU (so 5-10% for a 1-vCPU VM, 2.5-5% for a 2-vCPU VM, etc.), you've got a problem. I've seen VMs sit at 30%+ ready because a neighbor VM was pegging all cores. Quick check: look at the host's overall CPU usage in vCenter. If it's over 80% sustained, you're likely overcommitted.
5-Minute Fix: Identify the noisy neighbor
Same Performance tab, but now add "CPU Co-Stop" and "CPU Wait" for the same VM. If Co-Stop is high (above 2%), you're oversubscribing vCPUs to a single VM — don't give a VM 8 vCPUs if it only needs 2. If Wait is high, the VM's waiting on storage or network, not CPU. Real fix: look at the host's other VMs. Sort by CPU Usage in vCenter. Found one last month: a file server VM was chewing 90% CPU because its antivirus was scanning every file twice. Moved that VM to another host, and the slow VM's ready time dropped from 25% to 2% in minutes.
15-Minute Fix: Rebalance hosts and adjust shares
If you've got multiple hosts in a cluster, check vSphere DRS settings. Set DRS to "Fully Automated" with a migration threshold of 3 (aggressive) — it'll move VMs around to balance CPU load. But sometimes DRS needs a nudge. Right-click the cluster, select "Hosts," then "Cluster Resources" — you'll see a list of hosts and their CPU ready averages. If one host is sitting at 15% ready and others at 2%, manually migrate a few VMs off the hot host using vMotion.
Also check resource pools. I've seen admins set a high CPU reservation on one VM, starving the rest. Go to VM Properties > Resources > CPU — if you see a reservation that's more than 10% of the host's total, that's your culprit. Either reduce it or set a limit on the hungry VM. And if you're on older hardware (like a cluster of HP DL380 G8s), you might be hitting the physical limit of the CPU cores themselves. Upgrade to newer hosts with higher clock speeds or more cores. One client bought a three-host cluster when they really needed four — splitting VMs across four hosts cut their ready time by 80%.
Last resort: enable CPU Ready monitoring alarms in vCenter. Set a warning at 5% and a critical at 10% (per-vCPU). That way you catch it before users start complaining. But honestly, if you've done the above, you're already ahead of 90% of the fixes I've seen.