Hacker News Reader

Comment by edude03 | Hacker News Reader

I hate to be negative about progress but I wish companies would stop deploying DOCSIS and other time division multiple access strategies like that. Because DOCSIS is TDMA based its always going to have a problem with latency jitter and performance consistency and yet cable companies like to pretend it's the same as active Ethernet. Comparing 1gbit cable to 1gbit Ethernet is like comparing a dslr and a smartphone based on their megapixels.

View on HN

Progress isn't linear, they're working on all sorts of problems and they're not just deploying DOCSIS for the sake of it. This is the next best solution that Comcast and other ISPs can do with the current infrastructure they have. This is the right way of approaching the problem, deploy DOCSIS 3.1 as the next step and figure out the replacement for DOCSIS in the future. Customers want these bandwidth right now, not later.

Also, DOCSIS 3.1 has improvements for bufferbloat issue and it would stabilize the current latency issues. They're not ignoring it completely.

If we do what you suggest, we'd be stuck at 10mbps for a long time and instead, we're getting 100mbps bandwidth that does work for your typical usage of netflix, gaming, and so on in a family household. There's a limit we'll reach very soon where no one will care about more bandwidth but they will want rather stability/latency improvements. Data caps is the next biggest issue.

A lot of execs from DOCSIS community agrees that 3.1 is the last upgrade to DOCSIS standard and the next replacement would be a Fiber-based one, which will take a few decades to even figure out. I rather have DOCSIS 3.1 now and not in 10 years because you think we should not do this.

Our US congress had the chance to fix this back in 1990s and they screwed up with enforcing it properly. We could've had a better pure fiber-based national build out but it didn't work out.

"Our US congress had the chance to fix this back in 1990s and they screwed up with enforcing it properly. We could've had a better pure fiber-based national build out but it didn't work out."

Sounds a little similar to what happened in the UK, iirc. British Telecom wanted to do a full fibre deployment but the government upheld complaints of it being anti competitive, and now we have to endure copper *DSL being the default national broadband option and all the pitfalls that entails.

"Our US congress had the chance to fix this back in 1990s and they screwed up with enforcing it properly. We could've had a better pure fiber-based national build out but it didn't work out."

I'm not familiar enough with the industry to have an opinion on this, but it sounds very odd that a private corporation is so tied to government that it can't innovate on its own? How does this work? And how did it end up like this? I guess what I'm after is an ELI5 on this last part of your comment :)

You cannot break grounds without permits from the city, state, and federal governments (highway, streets, poles, etc) and then you must have an agreement from the land owners to lay the fiber through the rest to the places, think apartment buildings, rental places, and so on.

After that, you then have to schedule the construction to comply with the city regulations for quiet hours, cleanup regulations, inspections, and so on.

All of that cost a lot of money that only the federal government can handle alas the highway project we did.

For an example, we could have a federal law that said every time we rebuild the roads, there should be fiber laid in. Every new house should have new fiber by default, new water/gas pipes should have fibers laid in alongside of it, and so on.

IIRC, one city in CA already requires all new house constructions to have fiber wired up.

Exactly this. Fibre should be laid down to new housing just like Water and Gas Pipes. We used to do that with telephone wires, why not internet fibre?

Nationwide private deployment of fiber within the required parameters (i.e. universal service, network neutrality) is impossible without government intervention. The cost is simply too high and the created value is too hard to monetize.

Google Fiber?

Google Fiber doesn't deploy in any city that imposes build-out requirements, which compromises the universal service constraint. Build out requirements are also why my building and a couple of others have FIOS but nobody else does here in Baltimore. Verizon tried to build it, but the city wouldn't let it pick and choose which neighborhoods.

Google has also stated that Fiber is a way to shame ISP's into building fiber. It's not clear that an independent company that didn't have synergestic internet businesses would undertake the same capital investment.

Nearly every SP has made investments in "synergistic internet businesses." They appear to expect to extract the benefits of vertical integration in those businesses without making Google's investment in access.

Also, Google has announced the intention to bring Google Fiber to several major cities. It's not a demo, or hobby, anymore.

Switched ethernet is also practically TDMA. You can't send multiple packets in the same time. Although gigabit ethernet is full duplex, so you can receive and send in the same time.

Practically DOCSIS adds about 3-5 ms. Which is upper range of how much wifi adds latency.

Compare that to usual peripherals. Wireless mouses, about 50-100 ms latency, typical computer displays about 16-50 ms, flat screen TVs 80-200 ms.

Practically speaking modern Ethernet switches are non blocking though which allows every port to talk to every other port at full speed. With DOCSIS being TDMA there isn't a way to improve speed to the backbone (other than adding more channels / smaller time slots).

Also 3-5ms is probably at the low end since it heavily depends on the usage patterns of those on the same CMTS as you.

It's not TDMA, it's FDM. (It seems you've severely misunderstood DOCSIS.) And this is the best they can do with the existing network.

Feel free to run an Ethernet cable from my house to yours, though.

It's both TDMA and FDM. Your modem can bond multiple channels on the same coax but each channel is shared amongst all the subscribes on the CMTS. Each DOCSIS revision ups the spatial efficiency of each channel but also ups the number of channels.

On a side note, my ISP incentives users to upgrade their modems by offering more bandwidth for the same money if you buy a modem that can bond more channels. For example I had a 8/3 DOCSIS 3 modem on a 60mb plan, for the same price they gave me 100mbit for upgrading to a modem that could bond 24/8. It allows them to utilize the network more efficiently

But how big is the jitter? A bit of searching suggests it's typically under a millisecond, and that's barely anything. Most modems I've seen have a baseline latency of something like 20 milliseconds.

20ms baseline latency is pathetic. Ping from a data center in Los Angeles to one in San Francisco is typically a little over half what it apparently takes your modem to send a packet upstream. 20ms is like half way across the country. Why is latency at the last mile so awful?

What is the latency typically like on stuff like Google Fiber?

Baseline on my DOCSIS 3.0 modem is as low as 5 ms when I've tested it. That anyone should care about a 5-ish ms overhead is ridiculous imo.

It's great when it's that low, my problem with DOCSIS is that it's not consistent. When there is low utilization in your neighbourhood it's fine but when everyone is home for the holidays for example it becomes unbearable.

For example, I usually get about 8ms to a server hosted in the same city as me (10ms to google) but on bad days it gets as high as 60-80ms.

I regularly tether my LTE enabled phone where I get around 20ms but it's consistent.

The DOCSIS 3.0 modems we use at our local provider have no where near that much latency. 5-8ms is typical.

On my completely regular residental broadband, built with standard ethernet almost fifteen years ago, I have 4 ms to my server in the cloud (same country, different provider). If Google Fiber is built with similar parts it should be comparable.

Am on gigabit fiber here, next hop is <1ms as it should be.

> Comparing 1gbit cable to 1gbit Ethernet is like comparing a dslr and a smartphone based on their megapixels.

In both of those cases the majority of customers doesn't care about the difference.

People that play games and get 1ms latency vs 20-30ms latency within the same town/city care, and that isn't a very small market..

What do you propose they should do? Is it compatible with their existing infrastructure? Does it play nice with existing consumer devices? What does the roll-out cost and time-frame look like?

Reply on news.ycombinator.com