Full-Core PPPoE — every CPU core carries its share, even for PPPoE
across every CPU core
more subscribers, same tin
peak — better experience
no reboot, no new hardware
1 · The hidden bottleneck: one core does all the PPPoE work
Modern network cards are built to share load. They look at each connection — who it's from, who it's to — and hand it to a different CPU core, so a busy box spreads its work evenly and uses all of its cores. It works beautifully… for ordinary internet traffic.
PPPoE is different. On a fibre or DSL network, each subscriber's traffic is wrapped in a PPPoE envelope — and the network card can't read the address inside the envelope. Unable to tell one subscriber's flow from another, it does the only thing it can: it drops every PPPoE packet onto the same single core.
2 · The idea — open the envelope, share the load
BNGSOFT does in software the one thing the network card can't: it looks inside the PPPoE envelope, reads which subscriber flow each packet belongs to, and steers it to a fair share of every core — instantly, in the fast path, before any heavy work happens. The load the card couldn't spread, BNGSOFT spreads.
3 · What it delivers
- The bottleneck was never your real traffic — it was one pinned core. Remove it and the whole machine's CPU is available.
- Throughput scales with the core count you already own, not with a single core's ceiling.
- Grow your subscriber base on the same hardware — or consolidate onto fewer boxes.
- No single saturated core means no queue building behind it — less lag, less jitter exactly when the network is busiest.
- Smoother video calls, gaming and streaming for subscribers during the evening peak.
- Fewer “it's slow at 8 pm” support tickets rooted in a hidden CPU wall.
- Using the full box means fewer boxes for the same subscriber count — less capex, rack space, power and support.
- Defers the next hardware refresh: headroom you already paid for is unlocked in software.
- A software switch — no forklift, no new line cards.
4 · Proven on a live network
On a production BNGSOFT box carrying thousands of PPPoE subscribers, the PPPoE bottleneck was measured directly: the single “catch-all” path was carrying roughly 26× more traffic than a fair share — one core pinned while the rest coasted. With Full-Core PPPoE switched on, that load fanned out evenly across all twelve cores.
| On the live box | Before — card only | With Full-Core PPPoE |
|---|---|---|
| PPPoE load distribution | ~26× on one core, rest idle | ~1× — even across every core |
| Usable capacity | Capped by a single core | Full box — all cores working |
| Latency behaviour at peak | Builds behind the hot core | Smoother, no single wall |
| Packets dropped switching it on | — | Zero · seamless, live |
5 · Safe by design
Default-off, operator-controlled
Ships inert. You turn it on with a single setting when you're ready — and it stays on across restarts.
Fail-safe
If a core is ever momentarily unavailable, the packet simply follows the normal path — it is never dropped. There is no failure mode that loses traffic.
Order-preserving
Each subscriber's flow stays pinned to one core, so packets never arrive out of order — the network card's own guarantee, kept.
No new hardware, no reboot
Pure software in the BNGSOFT fast path. Enable and disable live; nothing to install, nothing to cable.
6 · Where it fits
Running out of headroom
One core near 100% at peak; the rest idle.
A hardware purchase looming to “add capacity.”
Outcome: unlock the box you own before buying another.
Evening lag complaints
Subscribers report stutter at 8–11 pm.
Box CPU “looks fine” on average — because it's one core.
Outcome: remove the hidden wall; smoother peak.
Fewer boxes, same subscribers
A fleet sized around a per-core PPPoE cap.
High rack, power and support overhead.
Outcome: pack more subscribers per box; shrink the fleet.
Full-Core PPPoE — stop paying for cores you can't use
PPPoE forces every subscriber onto one CPU core and caps a multi-core box at a fraction of its worth. BNGSOFT opens the PPPoE envelope, reads each subscriber flow, and fans the load evenly across every core — so you get more subscribers per box, steadier latency at peak, and a lower cost per subscriber, from the hardware you already own.
Zero-downtime. Default-off. Fail-safe. A software switch — no reboot, no new hardware.