Need Solution?
Contact us if you need a solution.
Subscribers, busy-hour bandwidth each and the frame size you actually carry go in. The packet rate, the line rate and the number of commodity servers that cover it come out. Every assumption behind the answer is written down below, so you can argue with it.
The figure that decides everything else. Take it from your own peak rather than from the package you sell: a gigabit tariff rarely means a gigabit at 21:00. A few megabits per subscriber at busy hour is a common starting point for a residential base.
Packet rate, not bandwidth, is what a forwarding engine runs out of. A smaller average frame means more packets for the same Gbit/s, which is why the calculator asks for it instead of assuming 1,500 bytes.
The answer is counted in servers you can buy from any vendor. Sizing one box beyond the arithmetic is what buys you a maintenance window and a failure domain you can afford to lose.
32 GB is the floor, 64 GB at 40G and 128 GB at 100G, with every memory channel populated so the controller runs at full width. An under-populated board costs more throughput than most people expect.
PPPoE and IPoE, dual-stack IPv6, hierarchical QoS per subscriber, RADIUS with change-of-authorisation. The vBNG / BRAS.
Port-block CGNAT with per-subscriber pressure sensing, plus NAT64 and MAP-T for IPv6-only access. XDP CGNAT.
Termination, translation, shaping and filtering are a single XDP program in the kernel, not four appliances in a row. How XDP forwards.
We would rather you measured it than believed it. A proof of concept runs on a server you already own.
Contact us if you need a solution.