Hardware Sizing Calculator⚓︎
The sizing calculator estimates how many Kubernetes nodes an xCore or xRouter user plane needs, and how large, to carry a target throughput and subscriber count while surviving the loss of one node. It fits user-plane pods onto each node after the fixed overheads, holds back the capacity the surviving nodes need to absorb a failed node's pods, and reports the throughput and subscriber capacity that remain guaranteed. Use it together with the sizing guide, which gives the reference node shapes and the storage figures.
Enter a throughput and a subscriber target; the smallest cluster that carries them with N-1 headroom opens below.
How to use it⚓︎
- Click Start the calculator, enter the target throughput in Gbps and the number of subscribers, choose kernel mode or VPP/DPDK for the user plane and the network scope, and click Calculate optimal configuration. The smallest cluster that meets both targets with N-1 headroom is shown; click Apply & Open Calculator to load it into the calculator.
- Adjust the number of nodes, the cores and memory per node and the cores per UPF pod with the sliders. The output metrics and the per-node headroom map follow every change.
- Click Simulate Node Failure to redistribute the last node's pods over the survivors. A red bar means a node would run out of CPU or memory; the cluster is not N-1 safe at that size.
- Click Download Spreadsheet for the configuration, the per-node breakdown and the model constants as CSV, or Print / Save PDF for a report. Import CSV reloads a saved spreadsheet.
What the model assumes⚓︎
| Assumption | Value |
|---|---|
| User-plane throughput per core | 0.5 Gbps in kernel mode, 2.5 Gbps with VPP/DPDK |
| Subscribers per UPF pod | 65,536 in kernel mode, 262,144 with VPP/DPDK |
| Memory per UPF pod | 2 GB plus 8 KB per subscriber; rounded up to whole 1 GB hugepages with VPP/DPDK |
| Steady-state CPU utilisation | 70% of the node, or less when the N-1 reserve needs more |
| Fixed overhead per node | 4 GB for the operating system, kubelet, CNI and monitoring |
| Control-plane nodes | The first three nodes also carry the SMF, database and cache (0.5 cores, 4 GB); a private or hybrid core adds the AMF, AUSF, UDR and UDM (2 cores, 6 GB) |
These are planning assumptions, not performance guarantees. Real throughput depends on the packet size mix, the network interfaces and the node hardware; validate a proposed cluster against the reference shapes in the sizing guide and with Expeto Support before ordering hardware.
Calculator⚓︎
The calculator opens with a default cluster; Reconfigure reopens the wizard with your targets.
Interactive cluster planner — models UPF pod migration, hugepage constraints, and latency-safe CPU utilisation.
CPU bars: Dark blue = active UPF cores | Light blue = failover-reserved headroom.
RAM bars: Dark blue = pre-reserved UPF memory | Light blue = RAM needed for migrated pods in failover.
Red = resource limit exceeded.