Isvara
·visits
Tools/04 · Cluster KPI

One red weighs eight greens.

A cluster is not the average of its VMs: score depth and breadth, and fold them into one KPI.

Isvara's reading of the guide, not an official Broadcom or VMware assessment.

Cluster KPI
yellow band
51/100
the guide’s engine
A plain mean would claim 79red is heard 8x louder than green
Depth · the worst
12
CPU ready · worst VM
Breadth · the average
88
CPU ready · cluster average
Start from a cluster
The signals100 = no contention · 0 = at the SLA breach

Each signal carries its own four-band score. Depth reads the worst VM; breadth reads the cluster average. The guide deliberately scores the average, not the % of VMs.

CPU readydepth vs breadth
depthWorst VM12×8
the deepest CPU queue any one VM suffers
breadthCluster average88×1
how widespread CPU queuing is
CPU co-stopdepth vs breadth
depthWorst VM70×2
multi-vCPU scheduling pain, at its worst
breadthCluster average92×1
co-stop across the population
RAM contentiondepth vs breadth
depthWorst VM85×1
the worst memory latency any VM sees
breadthCluster average95×1
memory pressure across the population
Also in the scoreone reading each
secondaryBallooned / swapped / zipped90×1
sticky reclamation signals; balloon gets 4x the threshold of swap+zip
depthDropped packets · worst ESXi90×1
host-level, not per-VM; green is deliberately not zero
leadingVMs being vMotioned (%)88×1
the cluster shuffling VMs to cope is a leading indicator
The override tablecan only pull the KPI down
overrideError packets · worst ESXi100
rare but severe, so it sits in a second table that can only pull the KPI down, never lift it
Weighted signals51
Error packets · worst ESXi100
Cluster KPI, the lower of the two51
Progressive weightagenormalised
green×17 now
yellow×21 now
orange×40 now
red×81 now
Who holds the vote now

CPU ready · worst VM is red, so it counts ×8: one signal of 9 holding 47% of the vote.

The book’s example: 1 perfect red among 8 perfect green scores 50, not 89.

What drags it mostKPI 51 · yellow
CPU ready · worst VM ×812
CPU co-stop · worst VM ×270
RAM contention · worst VM ×185
In the guidepart 3 · cluster performance

Depth reads the worst VM because averages arrive too late; breadth reads the cluster average because the population matters. Signals use the 20-second peaks so short bursts are not missed. Bands: green 75+, yellow 50+, orange 25+, red below 25.

The guide’s per-metric cluster thresholds live in its screenshots, not its text, so this simulator scores signals (0 to 100 each), not raw counter values.

Official Broadcom docstechdocs.broadcom.com

VCF Operations 9.1 documentation, each page verified. The guide notes the exact per-metric cluster thresholds are shown in its screenshots; the engine here (bands, weighting, override) is what the text specifies.