5 Reasons XDP-Based BNGs Outperform Traditional BNGs

Unlock Explosive Network Performance: 5 Reasons Why XDP-Based BNGs Are a Gamechanger | BNGSOFT

A traditional BNG is a chassis: line cards, a proprietary forwarding ASIC and a licence for every block of subscribers. An XDP-based BNG does the same job in software, inside the Linux kernel, on a commodity x86 server. Here are five concrete reasons that difference matters to an operator, and what to check before you believe any vendor, including us.

1. Carrier packet rates on hardware you already buy

XDP processes each packet in the network driver, before the kernel spends any work on it. On BNGSOFT’s data plane that gives 50 Mpps per server with PPPoE, CGNAT, hierarchical QoS and ACLs all enabled. Because packets per second is the limit, throughput follows frame size: 33.6 Gbps at 64-byte frames, about 110 Gbps at 256 bytes and about 144 Gbps at IMIX. A 1RU server with two dual-port E810 cards has 400 Gbit/s of port capacity, and the same software starts on a small 8-core box with dual 10G.

2. Every feature in one pass, with no feature penalty

On many platforms the headline number is measured with everything switched off, and each feature you enable takes a slice of it. In an XDP pipeline, termination, translation, shaping and filtering happen in the same pass over the packet, so the 50 Mpps figure is the number with the full feature set on. Translated packets never enter the Linux network stack, which is why full CGNAT in XDP uses 26 to 36 percent less CPU than kernel NAT for the same traffic.

3. It stays inside Linux instead of taking the NIC away

The other software route to high packet rates is kernel bypass, which hands the NIC to a user-space process polling it from dedicated cores. XDP keeps the interface in the kernel, uses CPU only when packets arrive, and passes anything it does not handle, such as ARP or control-plane traffic, to the normal stack. Routing daemons, monitoring and the debugging tools your team already knows keep working.

4. Upgrades that do not drop subscribers

Every eBPF program is checked by the kernel’s verifier before it loads, so a data-plane update cannot crash the machine the way a faulty kernel module can. Combined with state kept in shared maps, that makes live updates practical: in the OrionOS 7.05 lab run, a traffic-engine restart, a live upgrade and a kill -9 of the session service at 128,000 subscribers all ended with zero session flaps, the same Acct-Session-Id and no Accounting-Stop.

5. Fixes ship in days, not release trains

A chassis vendor fixes a security issue in its next software train, sometimes with a line-card refresh attached. A software BNG on a maintained kernel follows upstream: BNGSOFT moved from kernel 6.18.2 to 6.18.3 in five days. When a NIC driver misbehaves at line rate, the fix goes into the driver itself, rather than into a workaround around it.

What to check before you believe any of this

  • Ask for packets per second with every feature you need enabled, at a stated frame size, not Gbps with everything off.
  • Ask which NIC drivers run in native XDP mode. Generic mode works everywhere and is far slower; it is not a production fallback.
  • Ask what happens to sessions during an upgrade, and whether RADIUS sees an Accounting-Stop.
  • Ask how the licence is counted. A licence tied to a hardware model means paying again when you move to a bigger server.
  • Run a proof of concept on your own traffic. Frame-size mix decides the real Gbps number.

Go deeper

Go deeper

About us

Step into the vanguard of innovation – welcome to BNGSOFT. We’re not just another tech company; we’re the crossroads where cutting-edge solutions meet unparalleled excellence, designed with network architects in mind.

Recent Posts

Sign up for our Newsletter

Click edit button to change this text. Lorem ipsum dolor sit amet, consectetur adipiscing elit

Need Solution?

Contact us if you need a solution.