Comment by rayiner | Hacker News Reader

Comment by rayiner | Hacker News Reader

Parent Full thread

This is a little bit off. Looking at an actual comcast channel map: https://pickmymodem.com/ofdm-sub-carrier-ranges-on-docsis-3-.... The 24 DOCSIS 3.0 channels are about a gigabit in the aggregate, and the 94 MHZ OFDM channel is about another gigabit. Compared to 2.4 gbps down for GPON. (GPON is a decade older tech, so not surprising.)

There will be more people on a cable node than on a GPON (which is typically 16-32). But Comcast is N+0 on its DOCSIS 3.1 footprint (which is 75% of its network). That means they’ve removed amplifiers that extend the signal of the cable over distances, because those amplifiers get in the way of more aggressive techniques used in DOCSIS 3.1. That requires pushing fiber deeper into the network, and has the side effect of reducing how many subscribers are in each shared bus.

The upload limits are more an issue of legacy channel maps than insufficient bandwidth on the wire. The upstream/downstream portions are separated into different portions of the RF spectrum, and for historical reasons the split is heavily biased toward downstream. But symmetric DOCSIS is around the corner. There is also a lot of headroom left in DOCSIS 3.1. Compatible cable modems must support a minimum 192 MHz OFDM channels, which is almost 4 gbps. 10 gbps is possible if all the bandwidth on the coax is used for data rather than TV.

This is spot on. I do quite a bit of DOCSIS engineering work and this is a very good explanation.

It is going to vary quite a bit depending on your MSO and location but the "neighborhood" in modern DOCSIS deployments is getting smaller and smaller with Remote-PHY and N+1/N+0 deployments. What was traditionally 200+ customers on a node is shrinking down to much smaller numbers (dozens vs hundreds).

One guideline that is often thrown around is that the node will have at least twice the bandwidth of the highest speed tier and capacity is often added in excess of that based on usage and node size. You can also get fancy with dedicating channels to specic packages, so some of that bandwidth might only be "shared" with other customers on the same tier service as you.

Appreciate the information regarding specs. My observations are from repeated exposures on east and west coast of US to Comcast cable connections, which are noticeably slower in the evening when there is more network congestion, and have terrible latency. Various speed tests give me only a fraction of advertised bandwidth, and even then, it won’t be sustained. It feels like they’ve programmed the system to allow new connections to have a burst of bandwidth, and then taper it down.

I have yet to experience a coax internet experience as reliable and pleasant to use as a symmetric fiber connection. I presume it’s because Comcast is overselling their capacity, because they’re Comcast.

Your feeling is spot on. They definitely burst to a higher speed at the beginning of a transfer. What’s really amazing is that they’re proud of it! https://corporate.comcast.com/comcast-voices/what-is-powerbo...

Apple and Intel seem pretty proud of their burst speeds too. (And battery life, which is quoted based on running the CPU at short bursts and then going idle.)

And why shouldn’t they be? Consumer tech is properly optimized for burst usage. The network delivers the 10MB home page in a burst, then goes idle. The CPU bursts to 4 GHz to render it as fast as possible, then goes idle. The battery delivers 35W for half a second, then drops down to 5W. The whole system is designed around that usage profile.

The difference is that CPUs burst above the speed they can sustain. The alternative would be running slower all the time.

Comcast, in contrast, just lets you use more of their otherwise idle capacity, then throttles you. The alternative would be getting the full capacity of their network all the time.

Looking at the description in the article:

> This team discovered that at any given time there was some unused capacity in our network that could be used to temporarily enable faster connection speeds. PowerBoost was born.

TurboBoost notices there is additional thermal/power headroom, and boosts until that headroom is gone. PowerBoost seems very similar, except it's based on using unused capacity. The tapering is necessary, because PowerBoost uses up that excess capacity. The network tapers users off after the boost so there is excess capacity for other users to have a higher boost.

It’s only a zero-sum game when their network is at 100% capacity. If they were playing games to optimize usage at peak times I might not care (although they should increase capacity in that case) but when they do this at 5AM it’s clearly bullshit.

I don’t understand why you’re carrying Comcast’s water here. Nobody else does this nonsense. My FiOS connection is a gigabit all day every day. This isn’t a clever bit of network management that gives them an edge. It’s a clever bit of marketing which also allows them to cheat on speed tests.

“The tapering is necessary, because PowerBoost uses up that excess capacity.”

What the heck!? The network isn’t some kind of battery. There’s no buffer that gets filled up after 20MB on a TCP stream and they have to slow down to let it drain. I know you’re smarter than this.

> It’s only a zero-sum game when their network is at 100% capacity. If they were playing games to optimize usage at peak times I might not care (although they should increase capacity in that case) but when they do this at 5AM it’s clearly bullshit.

I don't know exactly how PowerBoost works. I'm going off the description in the article you linked. The article says that PowerBoost uses up excess capacity on the network when it's available. It follows that you can't PowerBoost everybody, even at non-peak times, without lowering the maximum amount of boost.

> I don’t understand why you’re carrying Comcast’s water here. Nobody else does this nonsense.

No offense, but to most people on HN the physical network is an abstraction that magically delivers bits from point A to point B, and they say crazy things like "networks shouldn't be oversubscribed." I spent my engineering career sitting around layers 2/3, so reading this stuff tends to put me into rant mode.

The article I linked is a PR piece. Don’t get technical details from it. I linked to it just to show how they’re proud of their bullshit.

Here’s some better info: https://www.dslreports.com/faq/14520

Relevant quote: “PowerBoost provides bursts for the first 20 MB downloading and the first 10 MB uploading of a file respectively on select Comcast High-Speed Internet services.”

Surely you admit that there is no sensible technical reason for a magic 20MB per-file limit on this.

I can think of one important technical reason: you can do a magic per-file limit without having global knowledge about network conditions. Imagine you're an edge router or switch. It's easy to do rate limiting--you just use a token bucket to drop packets above the rate the user pays for. It's also easy to just push packets into the core network as fast as you can, and let congestion control deeper in the network even things out.

Neither of those gets you the behavior you want, which is that even when the relevant route is at capacity, a user who starts a new connection will temporarily get a disproportionate share of the capacity compared to users who are doing downloading or streaming. A per-connection boost with a set limit approximates that behavior. Send packets as fast as you can for the first 20MB, then rate limit to the purchased bandwidth.

By contrast, how do you implement your proposal? You could dynamically allow each connection to pump packets as fast as it can unless the network was congested, in which case you transition to allowing each user a burst. But you don't really know the network is congested, especially at the edge, and particularly at layer 2, where this rate-limiting tends to happen.

Ok, so this is a highly intelligent and definitely user centric improvement to their network, and the only reason nobody else does it is because Comcast patented it. And the fact that the limit matches the amount of data used by popular speed testing services at the time is a total coincidence.

Cool, hooray for Comcast! I sure wish Verizon would take some lessons from them!

It would be one thing if it just helped on speed tests. But it also produced real-world benefits for things like web page downloads, MP3 downloads, PDF downloads, etc. It was an easy to implement hack that delivered real benefits for users. I don't see how it warrants so much sarcasm.

Again, I think this goes back to forgetting that networks involve real systems engineering. This is real infrastructure--some guy stringing wires on a poll or stuffing Ethernet switches onto a rack. If you're approaching it from the mindset that it's some abstraction where you can just turn a dial and give everyone a pony, yeah, I understand your position.

You apparently didn’t know about this thing before now, but you’re sure it helped with all this stuff? Why, because Comcast said so?

It warrants sarcasm because it was transparently obvious bullshit when it was announced, Comcast did some PR to spin it, and you’re falling for their PR hook, line, and sinker.

If this really was such a great thing for users then it would be commonplace, instead of being used by a single company that also happens to be one of the most hated companies in the country (and justifiably so).

You can’t just turn a dial and give everyone a pony, yet somehow non-Comcast providers manage to give better service. My FiOS connection is rock solid and delivers what it says with no speed testing games. Do you think I should ask Verizon to implement PowerBoost?

> You apparently didn’t know about this thing before now, but you’re sure it helped with all this stuff? Why, because Comcast said so?

No, because the benefits of getting a boost above rated speed for the first 20MB is obvious, as compared to giving everyone slightly higher speeds all the time.

> It warrants sarcasm because it was transparently obvious bullshit when it was announced, Comcast did some PR to spin it, and you’re falling for their PR hook, line, and sinker.

You haven’t explained why it doesn’t help real world use cases, or given me a technical explanation for how your proposed alternative could be implemented.

> If this really was such a great thing for users then it would be commonplace instead of being used by a single company that also happens to be one of the most hated companies in the country (and justifiably so).

“Comcast sucks therefore this must be bad” isn’t a valid argument. I spent an entire year on 4G LTE to avoid having to deal with Comcast’s customer service in Wilmington. That has nothing to do with the technical merits of PowerBoost.

> You can’t just turn a dial and give everyone a pony, yet somehow non-Comcast providers manage to give better service. My FiOS connection is rock solid and delivers what it says with no speed testing games. Do you think I should ask Verizon to implement PowerBoost?

FiOS has an underutilized network. For its history, it has struggled to get 30% uptake rate within its footprint. My apartment in Baltimore had FiOS and Comcast. I was the only one on my entire floor to ever subscribe to FiOS since the building was built 4-5 years before. I found out because apparently the ONT was never set up correctly. That’s great for customer experience, not so great for making money. And the making money part is important. Google pulled out of Fiber because it couldn’t make money. Verizon halted FiOS expansion. Comcast, as awful as it’s customer service might be, continues to make money and upgrade its network in an incremental way. Last time I had Comcast was probably 2007, in Atlanta. It was 20 Mbps with PowerBoost. Today they offer gigabit in my neighborhood. (I subscribe to Comcast’s fiber service, so I don’t use it, but many people choose it over FiOS.) Very few technology products have improved so much over that time period. Some, like OS X, have gotten worse. And that improvement didn’t happen by magic. You don’t just swap in a new CPU and double the speed. You pay linemen to go out and string fiber to split nodes. You pay them to take out amplifiers. You pay guys to swap gigabit switches for 10G switches and 10G switches for 100G switches. All that takes gobs of money. Comcast’s annual capex is hundreds of dollars a per per subscriber.

Is PowerBoost a trick that wouldn’t be necessary if you just dramatically over-provisioned the network? Sure. Is it nothing more than some PR stunt? Only if it doesn’t work technically, and you’ve provided no evidence of that.

Your position is “obvious” but mine needs technical evidence? OK then.

I'm just going to point out that in this one thread alone, I've read Rayiner write about individual DOCSIS versions, Comcast's channel map, OFDM, GPON, the performance differences between aggregated coax layer 2 standards and GPON, N+0, token-bucket rate limiting, congestion control, global network knowledge, FiOS utilization or lack thereof, a quantified argument of Comcast's rate of performance improvement compared to that of the entire technology industry, amplifier placement, linemen, and ISP CAPEX.

From you, I've read... something else.

By all means, keep going, because you're prying some pretty great comments out of Rayiner. I ran a network engineering group for a regional ISP full-time for a couple years and I didn't know any of this stuff. But don't kid yourself that you're the one making technical arguments; you'd be cheating yourself out of some pretty great information Rayiner's providing.

I’m not as knowledgeable about technical specifications as you are, but in my experience when I buy and use X Mb symmetric fiber connection, I actually get it all day every day, and I get it with sub 10ms or 5ms pings, and I have no problem streaming high def movies, multiple simultaneous FaceTime video calls, and downloads.

When I use Comcast (and I’m sure all other coax internet providers are the same, at least for residential users), what’s advertised is some nonsense like 200Mbit, but the upload is never mentioned because it sucks, and you only end up downloading things at a few Mbit anyway. Ping times are frequently in the hundred milliseconds, and multiple simultaneous streams/Facetime video/video games will slow everyone down.

In my mind, the 200Mbit Comcast is advertising is bullshit, at least since I know what a real 200Mbit up and down fiber connection feels like.

The last mile tech itself is largely irrelevant; Verizon FiOS is utter garbage where I live, and Comcast is reliable and fast. I know that’s heresy around here, but it’s not like Verizon is exactly any better than Comcast.

Comcast is offering 2 Gbps up AND down with supposedly no usage caps for $299/month in my area. Any idea what technology it uses?

Reply on news.ycombinator.com