Thanks! "OLT" was also new to me. In case others find it helpful:
> OLT = Optical Line Terminal.
> In ISP fiber (typically GPON/EPON) infrastructure, it’s the provider-side device at the central office/headend that terminates and controls the passive optical network: it connects upstream into the ISP’s aggregation/core network and downstream via fiber (through splitters) to many customers’ ONTs/ONUs, handling PON line control, provisioning, QoS, and traffic aggregation.
I have been worked for a regional ISP 10 years ago and having an architecture like that one, would be a godsend.
With centralized BNGs we were not able to apply upstream QoS policies for subscribers on the backhaulings and we had to apply policies on DSLAM access ports.
We ended using a couple of cheap Mikrotik as PPPoE concentrators for every access room, in a similar way as you did. But the reliability of Mikrotik routers was not the best
The code is mostly vibe coded and under the BSL. I think the interesting bit here is a single developer can write something like this with an agent. Does it make sense to open source such a thing or just each ISP write their own to their requirements?
I also don’t get the focus on handling DHCP renewals in the kernel fast path. With 2000 subscribers per OLT and say a 5 minute lease time that’s only a few renewals a second.
I'm curious as to what actually is the CPU <-> NPU bandwidth in these whitebox OLTs? Traditionally that has been sized for small amounts of punted control plane packets, then programming a fast path into the NPU for revenue traffic.
There's a commercial product available from 6WIND that makes this much more supportable for mission-critical networks. It leverages DPDK and delivers excellent performance at scale.
This is very elegant, but is treading some ground that for various reasons never got commercial traction.
- Cisco tried distributed BNG about ten years ago, their "cnBNG" running on their x86 UCS server line. See [0]
- A UK company called Metaswitch tried doing this with eBPF and some home-grown tech (VPP meets fd.io and special sauce) in about 2018. Interestingly they pivoted the tech to work on 5G where blazing fast user plane is essential [1]. They got bought by Microsoft, ground into glass, and wiped out five years later.
- There was a lot of talk in ~2020 about whether wireline (fiber) and wireless (5G cellular) infrastructures could converge, with the BNG becoming another node in the system, like an AGF, and authenticating users against a UDR. 5G was already very distributed and it made a lot of sense at least on market-techture slide decks.[2]
Looking back, the difficulty making this commercial was not splitting up the function, making it performant, or running it disagg on commodity hardware. The difficulty was finding a set of anchor customers who were experiencing such pain on their existing BNG that they would be prepared to jump ship from their big iron to something new knowing full well that the new system would only support 10% of what their old Lucent 7750s or Ericsson boxes could do.
Taking disagg as an example, it makes little sense unless your network is above a certain size. But if you run a big network, like DT or AT&T, say, then you will demand hundreds of features be present before you will look at an alternative. Does it work with my OSS. Does it support all the features of RFC XYZ and the special tweaks that only we have. Will it keep the three-letter agencies happy when they serve a warrant. Can it pass muster with my security people. Can the developer survive working with my procurement people long enough to make enough money to fund development.
No disruptive vendor --none-- has ever made it past this barrier into the network core, despite operators saying for years that they want to work with disruptors. That's why Nokia, Ericsson, and Huawei reign supreme and telcos haven't innovated in decades.
In ossified companies like telcos there's also the issue that the limitations of the existing equipment are being worked around with people. Those people derive their salaries from it, their manager derives his salary + prestige from managing such a headcount, and so on.
While the top brass might indeed be interested and benefit from more automation and a network that mostly runs itself, it's a bad deal for effectively everyone else in the company, so any attempts in that direction will never end up anywhere.
That's why legacy companies have been talking about "digital transformation" for decades now, yet it never progresses past simply digitizing the paperwork (and often creating more of it due to reduced friction), because enough people derive their job from said paperwork to make actual digital transformation politically untenable and impossible to deliver due to constant sabotage.
I mean, you see this with MikroTik all the time. The recent L3HW-enabled devices (up to 400G now) are so good it's crazy, and European onshore manufacturing, too. However, it doesn't support a subset of legacy "Enterprise" features, even though there's always a way to do the same thing using different architecture to how ISP guys have been trained many years ago, so instead we hear all the time that it's inadequate.
5G is a breath of fresh air in the sense that a lot of new techniques and broadly-applicable architectures were introduced to ISP's. I'm telling you, they HATED it. They absolutely hate learning new thing and that may as well be the largest blocker for disruptive players in the market.
Sounds like a really cool endeavour. I had no idea that ISP infrastructure was so heavily centralised. Hope the author succeeds in their quest to improve on this. I love that they're using simpler, cheaper hardware for this. Essentially, it sounds like it could reduce vendor lock-in for ISPs.
The [ONT → OLT(+BNG)] → Internet] sections of the paths will continue to be owned by commercial entities that can still be the subject of court orders and/or government pressure.
Even if you were to roll your own cable in the ground to your own ONT/OLT/BNG at some point you will need to acquire IP transit or peering from other commercial entities.
genpfault|1 month ago
Broadband Network Gateway (BNG)[1]
[1]: https://github.com/codelaboratoryltd/bng#bng-broadband-netwo...
bigwheels|1 month ago
> OLT = Optical Line Terminal.
> In ISP fiber (typically GPON/EPON) infrastructure, it’s the provider-side device at the central office/headend that terminates and controls the passive optical network: it connects upstream into the ISP’s aggregation/core network and downstream via fiber (through splitters) to many customers’ ONTs/ONUs, handling PON line control, provisioning, QoS, and traffic aggregation.
joshbaptiste|1 month ago
direwolf20|1 month ago
lormayna|1 month ago
We ended using a couple of cheap Mikrotik as PPPoE concentrators for every access room, in a similar way as you did. But the reliability of Mikrotik routers was not the best
westurner|1 month ago
From "Open Hardware Ethernet Switch project, part 1" https://news.ycombinator.com/item?id=43969857 :
> There are 48+2 port switches with OpenWRT support
Are there 48 port switches with 8 or more cores?
Edit:
From "Show HN: Spliff – Correlating XDP and TLS via eBPF (Building a Linux EDR)" (2026) https://news.ycombinator.com/item?id=46663446 :
> the current "Golden Thread" correlation architecture fundamentally requires userspace + kernel cooperation that can't be fully offloaded.
dontdoxxme|1 month ago
I also don’t get the focus on handling DHCP renewals in the kernel fast path. With 2000 subscribers per OLT and say a 5 minute lease time that’s only a few renewals a second.
esseph|1 month ago
binome|1 month ago
pharos92|1 month ago
https://www.6wind.com/vrouter-vsr-solutions/virtual-broadban...
kjellsbells|1 month ago
- Cisco tried distributed BNG about ten years ago, their "cnBNG" running on their x86 UCS server line. See [0]
- A UK company called Metaswitch tried doing this with eBPF and some home-grown tech (VPP meets fd.io and special sauce) in about 2018. Interestingly they pivoted the tech to work on 5G where blazing fast user plane is essential [1]. They got bought by Microsoft, ground into glass, and wiped out five years later.
- There was a lot of talk in ~2020 about whether wireline (fiber) and wireless (5G cellular) infrastructures could converge, with the BNG becoming another node in the system, like an AGF, and authenticating users against a UDR. 5G was already very distributed and it made a lot of sense at least on market-techture slide decks.[2]
Looking back, the difficulty making this commercial was not splitting up the function, making it performant, or running it disagg on commodity hardware. The difficulty was finding a set of anchor customers who were experiencing such pain on their existing BNG that they would be prepared to jump ship from their big iron to something new knowing full well that the new system would only support 10% of what their old Lucent 7750s or Ericsson boxes could do.
Taking disagg as an example, it makes little sense unless your network is above a certain size. But if you run a big network, like DT or AT&T, say, then you will demand hundreds of features be present before you will look at an alternative. Does it work with my OSS. Does it support all the features of RFC XYZ and the special tweaks that only we have. Will it keep the three-letter agencies happy when they serve a warrant. Can it pass muster with my security people. Can the developer survive working with my procurement people long enough to make enough money to fund development.
No disruptive vendor --none-- has ever made it past this barrier into the network core, despite operators saying for years that they want to work with disruptors. That's why Nokia, Ericsson, and Huawei reign supreme and telcos haven't innovated in decades.
[0] https://www.cisco.com/c/en/us/td/docs/routers/cnBNG/cnBNG-CP...
[1] https://www.linkedin.com/pulse/why-new-approach-pure-softwar...
[2] https://www.broadband-forum.org/pdfs/tr-470-2-0-0.pdf
Nextgrid|1 month ago
While the top brass might indeed be interested and benefit from more automation and a network that mostly runs itself, it's a bad deal for effectively everyone else in the company, so any attempts in that direction will never end up anywhere.
That's why legacy companies have been talking about "digital transformation" for decades now, yet it never progresses past simply digitizing the paperwork (and often creating more of it due to reduced friction), because enough people derive their job from said paperwork to make actual digital transformation politically untenable and impossible to deliver due to constant sabotage.
tucnak|1 month ago
5G is a breath of fresh air in the sense that a lot of new techniques and broadly-applicable architectures were introduced to ISP's. I'm telling you, they HATED it. They absolutely hate learning new thing and that may as well be the largest blocker for disruptive players in the market.
WhyNotHugo|1 month ago
Guestmodinfo|1 month ago
modernpacifist|1 month ago
Even if you were to roll your own cable in the ground to your own ONT/OLT/BNG at some point you will need to acquire IP transit or peering from other commercial entities.
direwolf20|1 month ago