Infinity Sentinel Enterprise | 3-Node High-Availability Cluster | 3x EPYC 9354

When every service runs on one host, one hardware failure stops everything until a replacement part arrives. Infinity Sentinel Enterprise spreads your virtual machines across three identical servers with shared Ceph storage. If one server fails, Proxmox VE high availability restarts its protected virtual machines on the other two, with typical detection and failover times of about two minutes according to Proxmox, and virtual machines can be moved off a server so it can be patched or serviced while they keep running. It is designed for organisations of roughly 250 to 1,000 or more staff, clinics, multi-department firms, municipalities, universities and research groups running 30 to 150 virtual machines on one site.

Each node is an ASRock Rack 1U12L4E-GENOA/2Q with a 32-core AMD EPYC 9354, 384GB of DDR5-4800 ECC registered memory on all twelve channels, two mirrored Kingston DC2000B enterprise NVMe boot drives, six Kingston DC600M 1.92TB enterprise SSDs with power-loss protection for Ceph, two 25GbE SFP28 ports, an Intel X550-T2 dual 10GBASE-T card for the cluster network, IPMI, and two 1000W hot-swap power supplies in 1+1 redundancy. Across the three nodes that is 96 cores, 1,152GB of memory and 18 Ceph storage drives, above Proxmox's recommended minimum of three nodes and 12 drives. With three-way replication the cluster holds 11.52TB usable, and we plan for about 80% fill.

The network is part of the design. Proxmox requires latency under 5 ms between nodes and recommends a dedicated network for cluster communication with a second link as a fallback, so each node's cluster traffic has its own 10GbE port, the second link uses the other port with the management VLAN, and Ceph and virtual-machine traffic run on the 25GbE ports. Two 25GbE switches with cables, 10GBASE-T switch ports (or a direct three-node connection) and about 2 kW of protected power on two or more circuits are quoted or supplied by you, and a site survey comes before the order.

Every node runs Proxmox VE 9 with a one-year Proxmox VE Standard subscription (10 tickets a year and a four-hour first response for critical requests); Proxmox requires all nodes in a cluster to use the same subscription level. Windows Server, SQL Server and other applications run as guests and are licensed separately, and because virtual machines can run on any node, Windows Server licensing is assessed for the whole cluster in your quote. Hyper-V failover cluster designs are available by quote (Storage Spaces Direct requires Windows Server Datacenter).

Compared with Sentinel Advanced, Enterprise uses the same chassis and processor but three servers instead of one, Ceph shared storage instead of local ZFS, a separate cluster network, six SSDs per node and a Standard subscription on every node. That matters for maintenance as well as failures: each node's drives sit in internal bays serviced with the server powered down, and in a cluster that becomes a planned, one-node-at-a-time task while the virtual machines run elsewhere.

Each node is assembled, its firmware aligned with the others and burned in for 72 hours. We then build the cluster, deploy Ceph with three replicas, configure HA groups and fencing, and power off a node under load to prove that its virtual machines restart on the others; the results go in your report, together with a documented node-maintenance procedure and a two-hour remote handover. Options include a fourth node, which lets Ceph restore full redundancy after a node loss, six more SSDs per node (23.04TB usable), 768GB per node, switching, a rack UPS, on-site installation and commissioning, monitoring, and a separate Proxmox Backup Server. Failover is a restart rather than zero downtime, one site is not disaster recovery, and Ceph replication is not a backup, so a separate backup target is required. Each node carries our 3-Year Hardware Warranty and Lifetime Technical Support. Custom-configured and quoted per project; built and tested in Ontario.