Latest / Logosofia no Dia a Dia / Why Your Fiber Upload Is Stuck at 250 Mbps
Transcript
- [Prompt] custom: You get very fast internet through a fiber optic system, it's very rare to get a symmetric connectio | Hosts: corn, herman [Response] Corn: Daniel sent us this one — he's on fiber in Israel, finally escaped the DSL dark ages, and he's noticing something weird. He's got a one gigabit connection, maybe two point five, maybe even five gigabits down. And yet the upload on all of these plans? Two hundred fifty megabits. Maybe as low as a hundred if you read the fine print. His question is basically: why? Is this a technical limitation, a business upsell strategy, or just a corner ISPs cut because most people don't notice? Herman: The answer is yes. But not in the way most people assume. Corn: That's the part that got me. You'd think fiber is fiber — light moves at the same speed both directions. And yet here we are, twenty twenty-six, and the average residential fiber plan in Israel is about as symmetric as a seesaw with an elephant on one end. Herman: Here's the thing that should immediately tell you this isn't an accident. Bezeq sells a five thousand megabit download plan. Two hundred fifty. They could make it symmetric. The hardware in the ground supports it. They choose not to. Corn: Daniel's instinct — that this might be an upsell play for business plans — he's not being paranoid. Herman: He's being observant. The same company that sells you five thousand down and two hundred fifty up for about a hundred forty nine shekels a month will happily sell you one thousand symmetric for three hundred ninety nine shekels. That's nearly triple the price for less total bandwidth. The only thing you're buying is the upload. Corn: Which means the bottleneck isn't the glass in the ground. It's the pricing sheet. Herman: And that's what we're going to unpack — the technical architecture that makes this possible, the business logic that makes it profitable, and what you can actually do about it if you're sitting there watching a YouTube upload crawl at a hundred megabits while your download could stream forty Netflix feeds simultaneously. Corn: Because here's the thing Daniel's really asking, whether he said it directly or not. If the technology can do symmetric, and other countries do symmetric, why is Israel still selling internet like it's two thousand and five cable? Herman: Let's put numbers on what "asymmetric" actually means for someone like Daniel. You're backing up a terabyte of footage to the cloud. On a five gigabit download plan, your upload is two hundred fifty megabits — and the SLA floor can drop to a hundred. That's not a rounding error. That's a twenty-to-one ratio on a good day, fifty-to-one on a bad one. Corn: It's not like the upload cap scales with the tier. A thousand megabit plan gives you two fifty up. Two point five gig? Two fifty up. Still two fifty. The download gets five times faster and the upload doesn't budge. Herman: Which is the tell. If this were a physics problem — if the glass literally couldn't handle symmetric throughput — you'd expect the upload to at least scale proportionally. The fact that it's a flat cap across every residential tier means someone set a knob and walked away. Corn: Walk me through the history here, because Daniel mentioned DSL was the reason he couldn't do cloud backups at all. Was DSL symmetric? Herman: Technically, ADSL — the A stands for asymmetric — was designed that way from the start. Twenty megabits down, maybe two up if you were lucky. But that was a genuine physics constraint. The copper pair could only carry so much, and the higher frequencies attenuated faster on the upstream. It wasn't a business decision, it was Shannon's law. Corn: Right, so nobody expected DSL to be symmetric. Then fiber arrives and the pitch is: finally, unlimited bandwidth both ways. And yet here we are. Herman: That's the puzzle Daniel's really pointing at. Fiber fixed the bandwidth starvation — you can actually upload a YouTube video now — but it reintroduced asymmetry at a completely different order of magnitude. DSL gave you twenty down and two up. Fiber gives you five thousand down and two fifty up. The absolute upload is a hundred times better, but the ratio is actually worse. Corn: Which brings us to the three explanations sitting on the table. Option one: the PON architecture physically can't do symmetric at scale. Option two: it can, but the cost of making it symmetric — the optics, the backhaul, the peering — eats into margins. Option three: they can do it, it's not even that expensive, but keeping upload scarce lets them sell business plans at triple the price. Herman: The uncomfortable truth — which we're going to get into — is that it's mostly option three, with enough of option two to give the ISPs a plausible technical alibi. Herman: Let's pop open the PON splitter and look at what's actually happening inside that fiber. PON stands for Passive Optical Network. The passive part matters — it means there's no powered equipment between the ISP's central office and your home. Just glass and a splitter. Corn: A splitter being literally a prism that divides one beam of light into sixty-four beams. Herman: The ISP has a device called an OLT — Optical Line Terminal — sitting in a data center somewhere. That OLT shoots a laser down a single fiber. Somewhere in your neighborhood, that fiber hits a passive splitter, which fans it out to up to sixty-four homes. Each home gets an ONT — Optical Network Terminal — the little box on your wall. And here's the key: all sixty-four homes share the same upstream wavelength. Corn: Share as in, only one can talk at a time? Herman: More precisely, the OLT allocates timeslots. It says ONT number one, you get to transmit for the next two microseconds. ONT number two, you're up. It's a round-robin, and if you're ONT number sixty-four, you wait your turn. That's fine when most people are just sending TCP acknowledgments and the occasional WhatsApp message. It's a disaster if four people on your splitter are all trying to upload 4K video simultaneously. Corn: The upstream is a shared party line, and the downstream is... Herman: The downstream is shared too, but differently. The OLT broadcasts everything to everyone — each ONT just decrypts the packets addressed to it. And the downstream wavelength on GPON runs at two point four eight eight gigabits per second. One point two four four. The standard itself was built asymmetric. Corn: That's the ITU-T G dot nine eight four spec, right? Herman: That's the one. GPON was ratified in two thousand three, and it baked asymmetry into the physical layer. Two point four eight eight down, one point two four four up, shared across thirty-two or sixty-four subscribers. Do the math on sixty-four homes sharing one point two four four gigabits — that's about nineteen megabits per home if everyone's uploading equally. Nobody designs for equal usage, so ISPs oversubscribe at ratios of fifty to one or higher. Corn: Which is how you get from a theoretical one point two four four gigabit shared pipe to an SLA floor of a hundred megabits per subscriber. Herman: That's on GPON. Now, XGS-PON — the ten gigabit standard, G dot nine eight zero seven point one — is symmetric by design. Ten gig down, ten gig up. The wavelength plan is different: fifteen seventy seven nanometers down, twelve seventy up. But here's where the alibi comes in. Even though the standard supports ten gig symmetric, ISPs configure it asymmetrically because they're still splitting that OLT port across sixty-four homes. Corn: The fiber itself can do ten gig both ways, but the moment you hang sixty-four houses off one port, symmetric ten gig to every home is mathematically impossible. Herman: If just four subscribers on that splitter tried to upload at one gigabit each, you'd saturate four gigabits of the ten gig upstream. Push it to ten users at one gig, and you're at the ceiling. Now imagine all sixty-four trying to upload at anything close to their plan speed. The OLT's upstream port would melt — figuratively speaking. So the ISP caps everyone at two fifty to keep the contention math from blowing up. Corn: That's the technical alibi you mentioned. They can point at the splitter ratio and say, look, we physically can't give everyone symmetric ten gig. Which is true. What they don't say is they could lower the split ratio. Herman: They could run thirty-two homes per port instead of sixty-four. They could run sixteen. Every time you halve the split ratio, you double the available bandwidth per subscriber. But that means running more fiber, installing more OLT ports, and spending more on the central office infrastructure. A sixty-four-way split is cheap. A sixteen-way split costs four times as much per home passed. Corn: The ONT on your wall — that's another cost lever. A symmetric XGS-PON ONT needs a burst-mode laser transmitter capable of ten gigabit upstream, plus bigger buffers to handle the timeslot scheduling. An asymmetric ONT that only does two point five gig up can use cheaper optics. Herman: The bill of materials difference is probably single-digit dollars at manufacturing scale, but multiply that across a few million subscribers and it's real money. Plus there's the backhaul. The ISP has to buy upstream transit from peering partners and Tier 1 providers. Historically, residential traffic has been about ninety percent downstream — Netflix, YouTube, web browsing. So ISPs built their transit contracts around that ratio. Buying symmetric backhaul capacity means paying for upstream bandwidth that sits idle most of the time. Corn: Which is a genuine cost, but also a self-fulfilling prophecy. If you cap everyone's upload, of course upstream usage stays low. Herman: And that's where we cross from technical constraint into business strategy — which is where this gets interesting. Corn: Let's follow the money. Bezeq's residential five gig plan — five thousand down, two fifty up — is a hundred forty nine shekels a month. Their symmetric business plan, one gig both ways, is three hundred ninety nine shekels. You're paying nearly triple for one-fifth the total bandwidth. The only thing you're buying is the upload. Herman: That's not a bug. That's the product segmentation working exactly as designed. The residential tier is priced for Netflix and WhatsApp. The business tier is priced for the moment you realize Netflix and WhatsApp isn't enough — when you're trying to push a hundred gigabyte Premiere project to the cloud and the progress bar says "eight hours remaining. Corn: Daniel mentioned cloud backups specifically. He said DSL made them impossible, but even on fiber, a full terabyte backup at two hundred fifty megabits takes about nine hours. At a hundred megabits — the SLA floor — it's over twenty-two hours. That's not "backup overnight." That's "hope nothing crashes for a full day. Herman: This is where the upsell psychology gets subtle. The five gig plan is the prestige tier — it's the one the ISP advertises, the one that makes you feel like you've got the fastest thing available. But the upload cap means it's actually worse value for anyone who does real work. You're paying for a download speed you'll never saturate, while the upload bottleneck makes certain tasks functionally impossible. Corn: Which nudges you toward the business plan. Not because you need a static IP or an SLA with a four-hour repair window — you probably don't — but because it's the only way to get upload speeds that make cloud workflows viable. Herman: The ISPs know this. The business plan isn't really for businesses. It's for the power user who's hit the residential upload ceiling and is willing to pay triple to remove it. The static IP and the SLA are just window dressing — they justify the price gap so it doesn't look like you're paying three hundred ninety nine shekels purely for upload speed. Corn: Even though you are. Herman: Even though you are. It's the same play airlines use with premium economy — the seat isn't that much better, but the legroom on regular economy is deliberately bad enough to make you consider it. Corn: Is this a scam? I don't think it rises to that. It's not fraudulent — the speeds are disclosed. But it's definitely what you'd call manufactured pain. The asymmetry is a choice, and the pain it causes is the sales funnel. Herman: The knock-on effect ripple out in ways most people don't think about. Take remote work. A VPN adds encryption overhead — figure ten to fifteen percent throughput loss. A video call in HD is about three to five megabits. A smart home camera uploading continuously is another two to five. Cloud sync for a family of four with phones backing up photos? That's bursty but can spike. Suddenly your two fifty up doesn't look like two fifty anymore. It looks like a contended, oversubscribed pipe that's being nibbled at from a dozen directions. Corn: If you're a small business — say a video editor or a photographer — you're not going to run your client delivery on a residential plan with a hundred megabit SLA floor. So you get pushed into the business tier, paying a hundred ten dollars a month for something that costs forty dollars in Tokyo. Herman: Let's talk about Tokyo, because the comparison is instructive. NTT's ten gig symmetric plan — not one gig, ten gig both ways — costs about five thousand eight hundred yen a month. That's roughly thirty-eight US dollars. For symmetric ten gig. Corn: Thirty-eight dollars. Meanwhile Bezeq's five thousand down, two fifty up is about forty-one dollars. Same price, fifty times less upload. Herman: It's not just Japan. South Korea, Singapore, parts of Scandinavia — symmetric gigabit or multi-gigabit residential fiber is unremarkable. So what's different? Part of it is density. Tokyo's population density means an OLT port might serve an apartment building with hundreds of units in a few hundred meters of fiber. Israel's suburban sprawl means longer fiber runs and fewer subscribers per splitter, which changes the economics. Corn: Density alone doesn't explain a fifty-to-one upload gap. There's a regulatory story here too. Herman: Japan's government effectively mandated fiber unbundling in the early two thousands — NTT had to open its infrastructure to competitors at regulated rates. That created price competition on speed, not just on marketing. Israel's wholesale mandate only arrived in twenty twenty-four, and it's still bottlenecked by Bezeq's OLT configuration. Partner and Cellcom can resell Bezeq's fiber, but they can't change how the splitter is provisioned. If Bezeq configures the OLT port for asymmetric service, the resellers are stuck with it. Corn: The wholesale mandate gave competitors access to the glass, but not to the configuration knobs that actually determine what the service looks like. Herman: It's like being allowed to sell tickets on a train but having no say over how many cars are attached. The Ministry of Communications opened the door, but they didn't mandate symmetric provisioning, so nobody offers it on residential plans — because the incumbent has no incentive to, and the competitors can't. Corn: Bezeq's monopoly hangover is real. For decades they were the only game in town for wired infrastructure. That history shapes how they think about product tiers — protect the high-margin business segment, keep residential asymmetric, and let the wholesale resellers fight over the low end. Herman: The original ARPANET protocols assumed peers could send and receive equally. The asymmetry we've normalized is entirely a commercial artifact of the broadband era — first with cable, then DSL, and now it's been carried forward into fiber even though the physical medium no longer demands it. Corn: What would actually force Israeli ISPs to offer symmetric residential fiber? Genuine infrastructure competition. If someone — say a new entrant or a municipality — started offering symmetric ten gig for fifty shekels more than Bezeq's asymmetric plan, the whole segmentation strategy collapses overnight. Herman: That's not hypothetical. It's happened in other markets. But in Israel, the barrier to entry for new fiber infrastructure is enormous — digging up streets, negotiating access to buildings, the whole permitting stack. So we're likely stuck with the current dynamic until either the regulator mandates symmetric provisioning on wholesale access, or a competitor finds a way to differentiate that isn't just price. Corn: Which brings us back to Daniel's original instinct. He asked if this is an upsell strategy for symmetric connectivity. The answer is yes — but it's an upsell strategy that works because the technical architecture gives it a plausible alibi, the regulatory framework doesn't prevent it, and the market hasn't produced a competitor willing to break the model. Corn: What can someone like Daniel actually do about this today, short of paying triple for a business plan? Herman: A few things, and they range from "call your ISP and complain" to "get creative with hardware." Let's start with the easiest one. If you're on XGS-PON — and most Israeli fiber deployments since about twenty twenty-three are — call your ISP and ask to be moved to a less congested splitter. The OLT port you're on might have sixty-three other homes hammering the upstream. Another port might have twelve. The ISP can reprovision you remotely. Corn: They'll actually do that if you ask? Herman: If you complain enough and use the right words. Say "contention ratio" and "SLA floor" and "upstream saturation." Sound like someone who might file a complaint with the Ministry of Communications. The squeaky wheel gets the less-congested splitter. It won't get you symmetric, but it might get you closer to that two fifty ceiling instead of the hundred megabit floor. Corn: That's the soft-power approach. What about the hardware hack? Herman: Bonding two connections. If you've got two separate fiber drops — or one fiber and a cable connection — you can use a load-balancing router that supports WAN bonding. Something like a Ubiquiti EdgeRouter or a MikroTik. Two asymmetric connections, each giving you two fifty up, can be bonded into a single logical pipe with close to five hundred megabits of upstream capacity. Corn: That costs what, another hundred forty nine shekels for the second line? Herman: So you're paying about three hundred shekels total for two asymmetric lines with bonded five hundred up, versus three ninety nine for the business plan with a thousand up. It's not perfectly equivalent, but if your upload needs are bursty — big file pushes a few times a day — it's a meaningful middle ground. Corn: For cloud backups specifically, Daniel's use case, there's a software angle too. Herman: Most backup tools default to full-file sync — they look at a changed file and re-upload the whole thing. Tools like Arq or Duplicati use block-level incremental backup. They only upload the chunks that actually changed. Edit a one-gigabyte video file and tweak two seconds of it? You're uploading maybe fifty megabytes instead of the whole thing. On a constrained upstream, that's the difference between a backup that finishes in minutes and one that runs all night. Corn: You're not fixing the asymmetry, you're making the asymmetry matter less. Herman: And that's the pragmatic playbook. But the bigger move — and this is where I think Daniel's question points — is demand-side pressure. Call your ISP and tell them you want symmetric residential fiber. Email the Ministry of Communications. The technology exists. XGS-PON can do ten gig symmetric. Twenty-five gig PON is coming in the next year or two, and it's symmetric by design. The only thing keeping upload capped at two fifty is a business model that treats upload as a premium feature. Corn: That model only survives as long as customers accept it. Herman: The internet wasn't designed as a broadcast medium. It was designed for peers. Every time you accept a fifty-to-one download-to-upload ratio, you're not getting a technical limitation — you're getting a pricing strategy dressed up in wavelength specs and splitter math. Corn: The glass can do it. The question is whether we demand it. Herman: That's the thing I keep coming back to. Twenty-five gig PON is right around the corner — the G dot nine eight zero four spec is finalized, commercial deployment expected in the next year or two. It's symmetric by design. Twenty-five gig down, twenty-five gig up, per wavelength. The physics is screaming at us that symmetry is the future. Corn: You're not optimistic. Herman: I'm not sure the business model hears the physics. What stops them from selling twenty-five gig down and two point five gig up, calling it "residential premium," and keeping symmetric twenty-five gig at a business tier that costs four times as much? The knob doesn't disappear just because the pipe gets bigger. It just moves. Corn: Right — the same playbook, new numbers. The ratio might even get worse. A ten-to-one asymmetry on a twenty-five gig pipe still leaves you with two point five up, and the ISP can point at the splitter math and say "sixty-four homes, do the division." The alibi scales perfectly. Herman: That's the open question. Does the sheer abundance of twenty-five gig finally make the segmentation absurd — like, at some point the upload is so fast even the capped version is more than anyone needs — or does the segmentation just get rebranded with bigger numbers and the same pain points? Corn: I suspect the latter, unless the market forces a change. Which brings us back to something you said earlier. The internet was built as a peer-to-peer network. Every node could talk to every other node as an equal. The asymmetry we've normalized — this whole idea that you're mostly a consumer of content, not a producer — that's not a law of nature. It's a story the broadband industry told us, and then built infrastructure to make true. Herman: It's a story that made sense in two thousand five, when the average person was browsing web pages and watching the occasional video. It makes no sense in a world where people are livestreaming, backing up terabytes to the cloud, running home servers, and working remotely over VPN. The usage patterns have shifted. The pricing hasn't. Corn: Daniel's question — is this an upsell strategy? — lands on a yes, but with an asterisk. The asterisk being that the upsell only works because the technical architecture and the regulatory landscape and the market structure all happen to align in its favor. Change any one of those and the whole thing gets harder to sustain. Herman: If this episode made you look at your internet bill and wonder what exactly you're paying for, maybe send it to a friend who's still on DSL and hasn't thought about upload speed since two thousand twelve. Corn: That friend needs to know what they're missing. And also what they're being denied. Herman: Hilbert's daily fun fact. Hilbert: A single surviving clay tablet from Copán, Honduras, dated to roughly 400 CE, contains a Mayan scribe's tally of tribute goods using a bar-and-dot numeral system where a shell-shaped glyph represented zero — one of the earliest known written zeroes in the Americas, predating its use in the Old World by centuries. Corn: A shell for zero. That's actually kind of beautiful. Herman: It really is. Corn: This has been My Weird Prompts. If you want more episodes that make you side-eye your utility bills, we're at my weird prompts dot com. Email the show at show at my weird prompts dot com. We'll be back soon. Herman: Until then, check your upload speed.