COMPARE

The benchmark everyone quotes has no subscribers in it.

Headline packet rates are measured forwarding bare packets with no sessions, no translation, no queuing and no filtering. That is not the workload a BNG runs. Here is how to compare platforms on the job they actually do.

6.9 Mpps

per core — VPP’s own published carrier-grade NAT figure, at 1,000 users

6.3×

below the 43.5 Mpps headline the industry quotes for the same stack

50 Mpps

per box with PPPoE, CGNAT, hQoS and ACLs all enabled

0

per-subscriber licence fees tied to hardware

Ask the questions that survive a proof of concept

Every one of these is answerable on your own traffic in an afternoon. If a vendor cannot answer them on your hardware, the datasheet is not the problem.

At what packet size?

A throughput figure without a frame size is not a figure. Packet rate is the limit; the bandwidth follows the average frame. Ask for both, and ask which features were on.

Read the brief →

With every feature enabled?

Termination, translation, shaping and filtering each cost cycles. Numbers measured with them switched off do not predict what happens when you turn them on.

Read the brief →

Kernel or userspace?

A userspace data plane takes packets away from the kernel and gives up its tooling and its drivers. Running inside the kernel keeps both — and the fix goes into the driver.

Read the brief →

What does a licence count?

Per-subscriber licensing tied to a chassis turns growth into a purchase order. Ask what the price does when subscribers double and when hardware is replaced.

Read the brief →

What happens on cutover?

Replacing a BNG is not a forwarding problem — the subscribers are already connected to the box you are removing. Ask how sessions move.

Read the brief →

Does it call home?

Procurement and security will ask whether the gateway depends on a vendor cloud, what it sends outbound, and what still works when the link to the vendor is cut.

Read the brief →

Head-to-head briefs

Written to be checked. Where a competitor published a number about themselves, that is the number used.

Settle it on your own hardware

Run the comparison yourself. A proof of concept runs on your own x86 against your own traffic — send subscriber count and busy-hour throughput and we size it with you.

Need Solution?

Contact us if you need a solution.