cluster.drgutman.org

kubertpmox

A second, separate Proxmox cluster: nine nodes, all-flash Ceph, and a deliberately lean layout. It exists so that disaster recovery does not depend on the cluster it is protecting.

9
nodes
~560
vCPU
~2.7 TB
RAM
13
Ceph OSDs
HEALTH_OK
Ceph status

Why a second cluster

Backups that live inside the system they protect are only half a backup. kubertpmox gives the lab an independent failure domain for VM backups, staging workloads, and disaster recovery. It is smaller and simpler than pmoxprime on purpose: fewer drive classes, fewer pools, fewer moving parts.

What is in it

The cluster runs Proxmox VE 9.1.2 across nine hosts. Ceph uses 13 OSDs — five NVMe and eight enterprise SSD — with a primary ceph_nvme pool at 3/2 replication. CephFS is active with two standby MDS daemons. Backups land on a Proxmox Backup Server on the lab network.

kubert

144 vCPU · 541 GiB
3× NVMe + 11× SSD · GPU
Four-socket compute heavy lifter. The node most likely to run a large restore or staging workload.

pmox2096

64 vCPU · 540 GiB
6× high-end SSD
Pure Ceph storage node: 4× 1.92 TB + 2× 3.84 TB enterprise SSD.

shinken

48 vCPU · 406 GiB
Ceph mon + mgr
Runs the Corosync ring and the monitoring role for the cluster.

pmox2093

48 vCPU · 406 GiB
Active Ceph mgr + NVMe
Active manager and NVMe OSD host.

pmox2072

56 vCPU · 135 GiB
19× SAS SSD, ZFS
Local ZFS storage, deliberately outside Ceph. 18× 1.2 TB SAS plus a 4 TB EVO.

pmox2070 / 2071

48–56 vCPU
135–270 GiB each
Ceph + VMs. Each contributes one NVMe and one SSD OSD.

Cluster summary

MetricValue
Nodes9 (PVE 9.1.2)
Total vCPU~560
Total RAM~2.7 TB
Ceph OSDs13 (5 NVMe + 8 SSD)
Ceph healthHEALTH_OK
CephFSActive, 2 standby MDS
Primary poolceph_nvme (3/2 replicated)
BackupsProxmox Backup Server on the lab network
CorosyncLab network

Drive detail

NodeNVMeSSDNotes
kubert3× 1 TB 970 EVO PlusMixed 1 TB EVO, 480 GB D2CSTK, 860 EVO, 4 TB EVOCompute heavy lifter
pmox2072—18× 1.2 TB ST1200MM0007 + 4 TB EVOZFS local, not Ceph
pmox2096—4× 1.92 TB + 2× 3.84 TB enterprise SSDCeph storage node
Standard nodes1–2× 1.88–2 TB NVMe1–2× Samsung EVOPaired OSD hosts

Snapshot date

These figures come from the September 2026 cluster survey. Hardware and pool state change; treat this as an architectural description, not live telemetry. The main cluster's status note is kept closer to current.

How it fits with pmoxprime

The two clusters are managed separately and do not share a Ceph quorum, which is the point. pmoxprime is the production-adjacent system with the capacity tiers and the research data. kubertpmox is the smaller all-flash system that can absorb a restore, host a staging copy, or keep services alive if the main cluster is unavailable.

Both clusters are connected to the same 10 GbE lab network and the same operational tooling. The benchmark work on this site focuses on pmoxprime because that is where the heterogeneous storage tiers are; kubertpmox is intentionally uniform.