Hypervisor metrics
Host pressure, from the hypervisor itself.
Declare the hosts your sessions land on and LoadGen reads them through a native hypervisor connection. No agent inside every guest, and the host figures sit on the same run as the user experience they explain.
Native connections
VMware vCenter and ESXi
A native connection to the hypervisor, not an agent on each guest.
Nutanix AHV
Through Prism Central or Prism Element.
KVM, via Prometheus
libvirt and KubeVirt exporters.
OpenShift Virtualization
The same Prometheus route as KVM.
Microsoft Hyper-V
Host performance counters.
In the score
Hypervisor is the sixth measure, and it only counts when you declare hosts.
The LoadGen Score is built from up to six measures. Hypervisor joins the other five when hypervisor hosts are declared for the run, and the weights re-normalise to the measures a run actually has, so a run without declared hosts is not penalised for missing one.
How the score is builtDeclare the hosts, once
The hosts a run lands on are part of the run configuration, so the connection is set up once rather than per test.
Live against your limits, then read in full
Host CPU, CPU ready, memory, storage latency and sessions per host are live during the run, against the limits you set. How many sessions a host carried, and per-step latency, are read on Test Analysis once the run has finished.
On the same timeline as SUT Monitoring
Hypervisor metrics come from the host. Windows performance counters bound through SUT Monitoring come from inside the guest. Both land on one timeline.
What a host figure does, and does not, tell you.
Host metrics tell you what happened on the hardware you tested, during the run you ran. They are measured evidence for a sizing conversation on that hardware.
A figure measured on one host, one image and one workload does not predict capacity on different hardware, a different image, or in production. Re-run it on the estate you are actually sizing.
- •Sessions, transactions, throughput, phases and errors are live during a run.
- •Host CPU, CPU ready, memory, storage latency and sessions per host are live too, against the limits you set, and a red banner names a host that goes over one.
- •Per-step latency and how many sessions a host carried are read in the analysis afterwards.
- •There is no live p95 and no live breaking point in the cockpit.
Point it at your own hosts.
We connect LoadGen to one of your hypervisors, run a workload against it, and walk through what the host figures do and do not justify.
