Daniel's been wrestling with this one for a while — he wants a fatter, more resilient internet pipe without signing up for a mystery box or building his own router from scratch. The term he finally landed on is SD-WAN, but what he actually cares about is the specific subset that does link aggregation and connection bonding. He wants a buyer's map with real numbers. Speedify, Bondix, Peplink SpeedFusion, hosted relays, the open-source end. And the core question is whether a middle ground exists — something where you're plainly renting a VPS as your aggregation endpoint while keeping control of your own local configuration. Who serves that, and what does it actually cost?
So let's start by untangling what SD-WAN actually means, because the industry has done a magnificent job of making that term mean everything and nothing.
The thing where a marketing category swallows the specific use case you actually want.
SD-WAN as an industry is about a five billion dollar category. It covers MPLS replacement, cloud-optimized routing, zero-touch provisioning, application-aware path selection — the whole enterprise WAN transformation pitch. Most of what gets sold as SD-WAN is policy-based routing. You say voice traffic goes over the fiber link, bulk downloads go over the cable link, and if one link dies the other takes over. That gives you resilience. It does not give you aggregation. Your Zoom call is still running over a single connection at any given moment.
And Daniel's after the thing where you actually combine them. Two ten-megabit connections behaving like a single twenty-megabit pipe, not two pipes with a traffic cop.
Right. True bonding splits individual packets or flows across multiple links simultaneously, and that requires a rendezvous point — a relay server somewhere — to reassemble everything in the right order before it hits the public internet. Otherwise the website you're talking to sees packets arriving from two different IP addresses and has no idea they belong to the same session. That relay is where all the cost and control tradeoffs live. Who runs it, where it sits, whether you can touch it.
So the spectrum is: at one end, black-box subscriptions where the relay is managed for you and you have no visibility. At the other, full DIY where you own everything but you're also the sysadmin. Daniel's asking whether anything lives in between.
Let's walk the field. Speedify first, because it's the most accessible entry point. Fourteen ninety-nine a month for an individual plan. What you're actually buying is a managed relay — Speedify runs the bonding servers, you install their client on your device. Windows, Mac, iOS, Android, Linux. The box sits on your laptop or your phone, not in your network.
So it's per-device, not per-network. That's an important distinction. If you've got three computers and a phone, you're installing it four times.
And each one bonds its own connections independently. The relay infrastructure is completely opaque — you can't swap in your own VPS, you can't tweak the bonding algorithm, you can't see what's happening inside their cloud. For fifteen dollars a month you get shared relay capacity, meaning you're contending with other users for the server's throughput. The hundred-twenty-dollar tier gives you a dedicated one-gigabit server with no contention.
Daniel's numbers check out on those. I pulled their pricing page.
They're accurate. And for a lot of people, that fifteen-dollar tier is genuinely the right answer. Install the app, it bonds whatever connections are available, you're done. The tradeoff is you control nothing about the infrastructure. It's a pure subscription.
Which is the black-box end of the spectrum. Now Bondix — this is the one that got interesting when I started digging.
Bondix is the most interesting player in this space precisely because they offer the middle ground explicitly. Their domain is bondix dot I O, by the way — bondix dot com is parked. Managed pricing: twenty-three euros a month for a hundred megabits per second of throughput, a hundred euros a month for one gigabit. They run the relay, you install their S.A.N.E. client.
S.A.N.E. being their proprietary bonding engine.
Yes. But here's the key. They also offer a self-managed option. Roughly three hundred ten euros per year, per gigabit tunnel, on your own VM. You rent their software license, spin up a VPS on DigitalOcean or Hetzner or whatever you like, and run their relay image. You control the local client config, you own the VPS, but the bonding logic is Bondix's proprietary code.
That's the closest thing to what Daniel's asking for. You can see the VPS, you know what it costs, you can pick the provider and the region. But you're not building the bonding engine yourself.
And that's the tradeoff. You're paying for their bonding protocol rather than using open-source Multipath TCP. Their hosted relays start around forty-five pounds a month if you don't want to manage the VPS yourself. But the self-managed tier — that's the middle ground. It exists. It's just not widely advertised because most of their customers go managed.
What about the bonding protocol itself? Speedify and Bondix both use proprietary engines rather than Multipath TCP. Does that matter?
It matters architecturally. Multipath TCP works at the kernel level — both the client and the relay need to support MPTCP, and it only bonds TCP traffic. Proprietary solutions like Speedify and Bondix work at the application or VPN layer, which means they can bond any traffic — UDP, TCP, whatever — but they lock you into their relay infrastructure. Bondix's self-managed option is interesting because it gives you the VPS control of open-source with the protocol flexibility of proprietary. If you trust their code.
Which is a real if. You're running their binary on a VPS you control. You can see the box but not what's inside it.
That's the tradeoff. Now Peplink is a different beast entirely. It's hardware-bundled. You buy a Peplink router — the Balance 20X starts around three hundred dollars — and SpeedFusion Connect is a feature you activate through a CarePlan subscription. The Standard tier, about fifty dollars a year, gives you up to fifty megabits per second of SpeedFusion throughput. Pro, about a hundred fifty a year, gives you up to two hundred megabits.
So three hundred dollars up front plus a hundred fifty a year for two hundred megabits. That's not nothing, but it's not enterprise-only either.
Right, but you're locked into Peplink's hardware and their SpeedFusion Cloud relay. You can also run FusionHub on your own VM as a relay endpoint, but that requires a separate license, and the pricing on that gets opaque fast. Peplink's real market is businesses with branch offices — the hardware approach makes sense when you're bonding connections across a whole office network rather than a single device.
And the two-hundred-megabit cap is worth flagging. Daniel mentioned it, and it's a real constraint. If you've got three hundred-megabit connections, your bonded tunnel still tops out at two hundred.
That's the CarePlan cap on aggregate SpeedFusion throughput, not per-link. It's an artificial ceiling built into the subscription tier. You want more, you pay more — and the Enterprise pricing isn't public.
The hosted-relay providers — Mushroom Networks and various white-label services — mostly resell Bondix or custom solutions. The pattern is consistent. Monthly fee for relay throughput, install a client, limited control. None of them offer the VPS-based middle ground as cleanly as Bondix's self-managed tier.
So that's the commercial landscape. Now the open-source end. OpenMPTCProuter, or OMR. This is full DIY. You install it on a Raspberry Pi or an x86 box, you rent a VPS — any provider, roughly five to fifteen euros a month for a basic one — and you run the OMR relay image on it. You control everything. The local router config, the VPS, the bonding algorithm. OMR uses Multipath TCP at the kernel level.
And it's free software. But you're paying for the VPS and your own labor.
The labor is the real cost. You're the sysadmin. No support, no polished UI, and you're responsible for keeping it running. If the VPS goes down or the kernel patch breaks on an update, you're the one fixing it at two in the morning. For a technical user, this is ideal — maximum control, maximum flexibility. But it's not a product. It's a project.
I've poked at OMR before. The value isn't just the kernel patch — it's the whole router OS and the community around it. But you're right, it's a project.
The architectural difference between OMR and the commercial options comes down to where the bonding happens. OMR bonds at the kernel level with MPTCP, which means both ends need to speak MPTCP. Commercial solutions bond at the VPN layer, which is more flexible — they can handle any traffic type — but proprietary. Bondix's self-managed option splits the difference. VPS control, proprietary bonding engine.
So does the middle ground exist properly? It sounds like yes, but it's basically Bondix self-managed or OMR on your own VPS. The difference is whether you pay for a polished bonding engine or do the work yourself.
That's the answer. Speedify and Peplink don't offer this — they're either fully managed or hardware-locked. The hosted-relay providers are reselling managed services. Bondix self-managed is the closest to what Daniel wants. About three hundred ten euros a year per gigabit tunnel on your own VM. You pick the VPS provider, you pick the region, you control the local config. The bonding engine is theirs, but the infrastructure is yours.
And OMR is the other side of that coin. Free engine, your VPS, your labor. The cost converges surprisingly close — a fifteen-euro VPS for a year is a hundred eighty euros, plus your time. Bondix self-managed is three hundred ten euros for the license, plus the VPS, but you're not debugging kernel patches.
The real question is whether the pricing makes this accessible to a non-business user. And I think the answer is yes, with a caveat. Speedify at fifteen dollars a month is the most accessible — install an app, done. Bondix managed at twenty-three euros a month is next, but requires more technical setup. Peplink is accessible if you're willing to buy hardware and a subscription. OMR is for enthusiasts.
The entry points are reasonable. The barrier isn't cost, it's technical comfort. The middle ground — Bondix self-managed, OMR — requires you to manage a VPS. That's a non-starter for most people. Not because they can't afford it, but because they don't want to learn how to SSH into a server and keep it patched.
And that's the thing the industry doesn't say out loud. The relay is always the bottleneck, and the pricing always obscures it. Speedify is actually the most honest pricing in this space — a flat fee for a known quantity. You know what you're getting, even if you can't see inside the box.
Bondix's self-managed tier sounds cheap until you add up the VPS cost and your own time. Three hundred ten euros a year for the license, plus maybe a hundred eighty for the VPS, plus the hours you spend configuring it. Suddenly you're at five hundred euros and a weekend of work.
Versus a hundred twenty dollars a month for Speedify's dedicated server — fourteen hundred a year — but zero setup time. The math depends on what your time is worth.
Peplink's pricing is misleading in a different way. The two-hundred-megabit cap means you can't exceed that even with faster individual links. If you've got three hundred-megabit connections, you're leaving a hundred megabits on the table. The CarePlan is cheap, but it's rate-limited in a way that isn't obvious from the marketing.
And if you want to use your own VPS as the relay with Peplink, you need a FusionHub license, and that pricing isn't on the website. You have to talk to sales. Which tells you who they're selling to.
The other thing that doesn't get talked about enough — VPS providers and sustained throughput. You mentioned this once. The real bottleneck for VPS bonding is often the provider's tolerance for you pushing a gigabit through their network twenty-four seven, not the software.
That's right. A fifteen-euro VPS on Hetzner or DigitalOcean will give you a certain amount of bandwidth, but if you're saturating it continuously, you might hit hidden caps or get a polite email from the abuse team. The VPS cost for bonding isn't just the listed price — it's finding a provider that won't throttle you or kick you off for using what you paid for.
Which is another argument for the managed services. Speedify and Bondix have already solved that problem on their end. You're paying them to deal with the upstream providers.
Let me address one more thing before we move on. The misconception that SD-WAN equals connection bonding. Most SD-WAN solutions do policy-based routing, not packet-level aggregation. They'll send your VoIP over link A and your file transfer over link B, and fail over if one link dies. That's useful. It's not bonding. Bonding requires that relay server to reassemble packets. If a product doesn't have a relay component — a cloud endpoint or a VPS you configure — it's not doing true bonding.
And some solutions claim peer-to-peer bonding without a relay. In practice, almost all of them use a cloud relay for packet reassembly. The physics of it — two different IP addresses talking to one website — means something has to sit in the middle and make those look like one connection.
Unless you control both ends of the conversation, which is not the use case Daniel's asking about. He wants to bond his home connections and talk to the regular internet.
Right. So to put the buyer's map on the table. Speedify: fifteen dollars a month, managed relay, per-device, zero infrastructure control. Bondix managed: twenty-three euros a month for a hundred megabits, managed relay, more technical setup. Bondix self-managed: about three hundred ten euros a year per gigabit tunnel on your own VPS — that's the middle ground. Peplink: three hundred dollars for hardware plus a hundred fifty a year for two hundred megabits, hardware-locked. OMR: free software, five to fifteen euros a month for a VPS, full control, full responsibility.
And the hosted-relay providers are mostly reselling Bondix or custom solutions at forty-five pounds a month and up. They're not a separate category, they're a distribution channel.
The middle ground exists. It's Bondix self-managed or OMR. The question is whether you want to pay for the bonding engine or build it yourself.
There's a deeper architectural point here about where the bonding happens. Speedify bonds at the device level — your laptop is the bonding endpoint. Peplink bonds at the network level — the router is the endpoint, and everything behind it benefits. OMR can do either, depending on how you set it up. Bondix is device-level with their client, but their self-managed option lets you put the relay wherever you want.
For a home user with multiple devices, network-level bonding is almost always better. You don't want to install a client on every laptop and phone. You want the router to handle it.
Which pushes you toward Peplink or OMR on a router box. Speedify and Bondix are fundamentally per-device unless you set up a dedicated bonding machine and route everything through it — which, at that point, you're building an OMR-style setup anyway.
So the buyer's map also depends on how many devices you're trying to bond. Single laptop? Speedify, done. Whole house? You're looking at Peplink hardware or OMR on a Pi.
Or a small x86 box running Bondix as a gateway. But that's getting into advanced territory.
Before we get to Hilbert, one more question. Is any of this actually usable for a normal person?
Speedify is. Install the app, it finds your connections, you're bonded. Everything else requires some degree of technical comfort. Bondix managed is close — their client is straightforward — but you still need to understand what you're bonding and why. Peplink requires you to configure a router. OMR requires you to be comfortable with Linux.
So the answer is yes, it's usable, but the complexity scales with control. The middle ground Daniel wants — VPS-based, local control — is accessible to anyone who can manage a VPS. That's a bigger pool than "can compile a kernel" but smaller than "can install an app."
That pool is growing. More people are comfortable with VPSes than five years ago. DigitalOcean and Linode have made it approachable. But it's still a technical audience.
The pricing is not the barrier. Fifteen dollars a month for Speedify, twenty-three euros for Bondix, three hundred up front for Peplink — those are consumer-level prices. The barrier is knowing what you're buying and why.
Which is why we're having this conversation.
Hilbert: Three hundred ten euros a year sounds cheap until you're the one SSHing into a VPS at eleven at night because the tunnel dropped and your wife is asking why Netflix is buffering.
That's the labor cost.
Hilbert: I ran a small ISP in rural Vermont in the late nineties. Bonded two fifty-six-k modems with a custom Linux script. Called it the Vermont Double-Wide. Worked about thirty percent of the time. The script is still on a floppy disk in my office.
Does the floppy still read?
Hilbert: Haven't tried. The modem drivers probably don't exist anymore. But the fundamental problem hasn't changed in twenty-five years. The relay is always the bottleneck, and the pricing always hides where the bottleneck actually is.
Say more about that.
Hilbert: Speedify's fifteen dollars is the most honest number on that list. You pay, it works, you don't think about it. Bondix's self-managed tier — three hundred ten euros a year, plus the VPS, plus your time — that number looks precise but it's not the real cost. The real cost is the hour you spend debugging a routing table at ten PM. Peplink's two-hundred-megabit cap is the same kind of hidden thing. The number on the box says two hundred, but what it means is you can't ever go faster than that no matter what you plug in.
The CarePlan pricing makes it look cheap until you realize the cap is aggregate.
Hilbert: Fifty dollars a year for fifty megabits. That's a dollar per megabit per year, roughly. Sounds fine. But if you've got three hundred-megabit connections, you're paying for capacity you can't use. The math only works if your individual links are slower than the cap.
Which for most home users, they probably are. A typical cable connection is a hundred, maybe two hundred megabits.
Hilbert: For now. But the cap doesn't move when your connections get faster. That's the trap.
The thing I keep coming back to with Bondix self-managed — you're running their binary on your VPS. You can see the box, you can pick the provider, but the bonding engine is a black box inside a transparent box.
Hilbert: That's not nothing, though. If the VPS provider throttles you, you can move. If Bondix goes down, you still have your VPS and your config. With Speedify, if they go down, you have nothing.
Portability matters.
Hilbert: The floppy disk still works because it's just a script. It doesn't phone home to a license server. If I wanted to run the Vermont Double-Wide tomorrow, I could — assuming I could find a machine with a floppy drive and two modem ports. The point is, I own it. Bondix self-managed is closer to that than anything else on the list, except OMR.
OMR you own. The code is on GitHub.
Hilbert: The cost is your weekend. I'm not saying one is better. I'm saying the pricing sheet doesn't tell you which kind of cost you're signing up for.
That's the distinction between money and control. Speedify is all money, no control. OMR is all control, no money — but your time is the currency. Bondix self-managed splits both down the middle.
Hilbert: The thing nobody advertises is that bonding two connections never gives you the sum of their speeds. You lose some to overhead — packet headers, reordering, the relay's processing time. My two fifty-six-k modems theoretically gave me a hundred twelve kilobits. In practice I got about eighty on a good day. The same math applies to gigabit connections, just with bigger numbers.
Overhead eats ten to twenty percent, depending on the protocol.
Hilbert: At least. So when Bondix says a hundred megabits, they mean after overhead, or they mean the tunnel capacity before overhead? The pricing page doesn't say. Speedify doesn't say either. Peplink's cap is a hard limit regardless of overhead. Nobody wants to put that on the brochure.
That's a real question. If the hundred-megabit tier is tunnel capacity, your actual throughput is more like eighty. If it's throughput, the tunnel is running hotter than a hundred.
Hilbert: I'd assume tunnel capacity until proven otherwise. That's how these things usually work.
The real numbers are worse than the brochure.
Hilbert: The real numbers are always worse than the brochure. That's the first thing I learned running an ISP. The second thing was that nobody cares as long as it works most of the time.
That's the pitch for Speedify, honestly. It works most of the time, and you don't have to know why.
Hilbert: Which is worth fifteen dollars a month to a lot of people. Including me, probably, if I didn't already have the floppy disk.
You're not actually going to run the Vermont Double-Wide again.
Hilbert: No. But I like knowing I could.
If you take one thing from this conversation, it's that the middle ground exists, but it's not a product category — it's a specific tier from one company and a project from the open-source community. Bondix self-managed, about three hundred ten euros a year on your own VPS, or OpenMPTCProuter, free but demanding. Everything else is either a managed black box or a hardware bundle.
The thing that changed how I see this — the pricing is consumer-accessible across the board, but the real cost isn't on any pricing page. It's the hour you spend debugging a tunnel at midnight, or the cap you didn't notice until you tried to push past it.
One open question before we go. As more ISPs deploy IPv6 and carrier-grade NAT, does bonding get harder or easier? The relay needs to handle NAT traversal, and some bonding protocols struggle with double NAT. If your connections are behind two layers of CGNAT, the relay might not be able to reach you at all.
That's a real concern, and it's one of the things the commercial protocols handle better than raw MPTCP — they've built NAT traversal into their clients. But the rise of five-gig fixed wireless access could make bonding more relevant, not less. If you have two five-gig connections from different carriers, bonding them could rival fiber without the trenching cost. The pricing models just haven't caught up to that possibility yet.
Thanks to Hilbert Flumingtop for producing, and for the floppy disk we will apparently never verify.
This has been My Weird Prompts. If you want to hear more about the packet-level mechanics of bonding or the history of Multipath TCP, those are deep dives we've done — but you don't need them to follow this one. Leave a review if you want a step-by-step on setting up Bondix self-managed or OMR.
We'll be back soon.