Google Fiber Changes Server Ban Language in TOS
You might recall that Google Fiber took a lot of criticism back in July for language in their terms of service that technically bans servers. The criticism originated with a user who filed a complaint with the FCC
, claiming that Google's language aimed to prohibit Google Fiber users from doing commonplace things like running Minecraft servers or using Slingboxes.
As noted at the time
, the hysteria was a little overblown in that nearly all ISPs contain this language to protect themselves from extremely heavy users trying to run a business on a residential line, and in many instances the language is never even enforced. Google primarily took heat because they professed to be offering a service that's different from the status quo, and the status quo is rights-eroding and ambiguous fine print.
Fast forward to last week, when Google Fiber appears to have quietly updated their terms of service and acceptable use policy
with new language that clarifies that they're not interested in prohibiting most reasonable uses of the connection:
However, personal, non-commercial use of servers that complies with this AUP is acceptable, including using virtual private networks (VPN) to access services in your home and using hardware or applications that include server capabilities for uses like multi-player gaming, video-conferencing, and home security.
As with most ISPs, you'll likely only gain the attention of Google Fiber's engineer team if you're pushing numerous terabytes monthly as you attempt to run a porn and poker empire out of your bedroom closet.
Re: It is sad they resisted this change As cheap as bandwidth is, it's not as cheap as Google is selling it for. They are betting on the fact that people won't utilize even a fraction of it and it is a very safe bet, as long as they can keep people from hosting high traffic servers.
Re: It is sad they resisted this change
said by silbaco:The problem is that nothing is charged per gigabyte online. Google pays for a backbone and can max that out while paying a single fixed cost.
it's not as cheap as Google is selling it for.
The cost is the price of google's backbone bandwidth divided by its total paying customers. Google is not charged more if customers max out the backbone. Google is only charged more if they get more backbone bandwidth to deal with total customer load. But it is a fixed cost increase.
If google oversells 10:1 and thus only has 100mbit of dedicated bandwidth per user that is guaranteed, then they are not going to have issues. They would just need to implement QoS that can cap max speeds to handle congestion. People being throttled down to something between 1 gigabit and 100 megabits probably aren't going to notice.
But we have no idea what ratio google will use when they oversell their backbone.
Wisconsin Rapids, WI
Re: It is sad they resisted this change The truck roll and equipment probably costs more than $300, then you need to include maintenance and customer care.
I think the idea was to offer "free" because of their deal with the city, but they needed a bit of a hurdle to not get trolled.
I don't have Google fiber, but I do have fiber and it involved a 3 person team to come out and trench the fiber from the curb to my house which was about 30 minutes, then another person to connect the fiber from the curb to the newly laid fiber to my house, which was another 20 minutes, then a 2 man team to get the fiber into my house and hook things up.
That's about 4 man hours of work just to Internet into the basement. Assume $15/hour, and that's $60, then add in equipment costs, rental fees, and fuel, probably another $100-$200, then the modem+UPS, another $150-$250.
Ballparking here, but my guess is $400.
I'm not even including other things like port usage back at the ISP or electrical costs or any other random thing.
$300 is very much likely to be a loss just for day one, yet alone the next 7 years.
Wisconsin Rapids, WI
Re: It is sad they resisted this change The cheapest bandwidth can cost is the cost of the ports. The real question is how much do the ports cost and how many ports does Google need.
Local bandwidth is easy. 1gb fiber tech is cheap when talking about 40km-80km ranges, but at some point you need to aggregate all of these customers and push them over your trunk.
The trunk is where things get expensive. Tech to transport 100gb/s over potentially thousands of kilometers is not cheap. What Google is banking on is that even though they're offering 1gb speed, very few people will make use of it.
As expensive as trunk bandwidth and transit is, the majority of the cost for most ISPs is not the trunk, but the last mile. Actual trunk demand is relatively low because most customers don't use their connections that often. Even a few high bandwidth users won't make much of a difference.
While a few heavy users make up the bulk of the total data moved, they make up a very small piece of peak bandwidth used. Since congestion and costs are entirely based on peak usage, heavy users really don't make a difference.
Well, they did back before YouTube and Netflix. P2P was so much more bandwidth than web browsing, but P2P is smaller than media streaming.
Case studies actually agree with this. On average, going from 10mb/s to 1gb/s only increases trunk load by about 30%. As long as this remains true, Google can naturally upgrade their trunk as long-haul tech gets cheaper, while not having to touch their last-mile infrastructure for a long time.
Now if a new wonderful high bandwidth and high demand service popped up over-night, Google would be screwed. But based on lots of research of how services and demand develop, this risk is phenomenally small.
| |said by silbaco:No they are not. Every single sub gets a dedicated single mode fiber. The bottle neck is literally the backbone.
As cheap as bandwidth is, it's not as cheap as Google is selling it for. They are betting on the fact that people won't utilize even a fraction of it and it is a very safe bet, as long as they can keep people from hosting high traffic servers.
Re: It is sad they resisted this change
said by DataRiker:I wouldn't assume that necessarily. I haven't seen any concrete specifics as far as how they deploy their service, other than the fact that they're operating some sort of WDM-PON & Active Ethernet hybrid setup, essentially dedicated virtual fiber strands in the form of wavelengths over shared fiber, all coming back to ordinary Ethernet switches (configured with all SFP ports with SFP+ uplinks).
No they are not. Every single sub gets a dedicated single mode fiber. The bottle neck is literally the backbone.
I don't see Google deploying a completely non-blocking infrastructure for a residential service. If they are, I doubt they're making much money. More than likely they're running everything off of 1gbps ports with 10gbps uplinks, meaning either 24 or 48 homes are sharing a single 10gbps uplink.
There are stackable (allowing for non-blocking) Active Ethernet solutions out there. Unfortunately they're quite expensive, as usually you are forced to buy environmentally-hardened equipment. It's my understanding that everything is homerun back to a fiber hut, which is more than likely environmentally controlled, not requiring more expensive gear.
Wisconsin Rapids, WI
Re: It is sad they resisted this change At one point they described their setup as a single chassis able to handle up-to 500 customers with up-to 400gb-500gb of uplink.
Other commercial avaliable and popular GPON linecards have 8 GPON ports, which are 2.5gb/s per port down and 1.25gb/s up, and each line card has 2 10gb ports, which perfectly matches the peak bandwidth in any one direction.
Near-non-blocking setups, are what's encouraged. There is nothing stopping someone from not using those 10gb ports, but they've already paid for them.
The other cool thing is they can be used for redundancy. So say you can have 20 line cards in a chassis, each with 20gb of uplink. If any one uplink fails for a given linecard, it can fail-over to sharing another linecard's uplink.
This does have to be setup, but it can be done and is the recommend setup based on documentation. Typically they have linecard pairs, so in normal mode, you have 1:1 for the uplink, but if a linecard's uplinks both fail, then the paired linecard for failover, now has to share, with a 2:1.
You can also get linecards without uplinks then get dedicated linecards for just uplinks. So say 18 linecards with 8 GPON ports each, and 2 linecards, both with 2x100gb ports, giving 400gb/s to be shared with 144 GPON ports, which gives an under-subscribed ratio of 0.9:1
Now I get a bit fanciful, as I don't know how much these core routers actually cost.
Take your chassis with 400 customers, each with 1gb over WDM-GPON, give it 400gb/s of uplink via 4x100gb ports, then plug that into one of those core routers with 30tb/s of non-blocking routing via 150x100gb ports.
To keep things symmetrical, every one port plugging into a chassis, needs another plugging into the trunk. So only 75 ports can be used as uplinks for the chassis. That means 75,000 customers can have 1gb of non-blocking local routing.
But wait, their's more!
Newest commercial available tech can multiplex 80x100gb links over a single pair of fiber. Since you're only using 75 ports, you can support all 75,000 1gb customers with not only non-blocking access to the trunk, but non-blocking access ON the trunk.
If you think that's crazy, you should see the 1pb/s routers that can support 500 1tb/s ports(when they come out, right now only 400gb/s ports) and a 144 strand fiber cable that will allow 576tb/s of peak data transfer using the above multi-plexing of 100gb ports.
So two fiber cables, both the thickness of a pencil, can support about 500,000 customers with full non-blocking trunk access.
I'm sure the cost of this setup would be insanely high, but it can be done.
Statistics shows that we don't need non-blocking speeds up-stream. It is a statistical near-impossibility for everyone of a large group to use all of their bandwidth at the same time.
As long as you can keep the "nodes" from congestion, then you're good to go. You only need non-blocking speeds at the smaller levels of aggregation, like the ports, or maybe the uplinks of a chassis.
Re: It is sad they resisted this change
said by brad:I never said it makes sense. Caps are the worst way to prevent peak congestion. But that is what cable companies implemented. They probably didn't have any QoS capability at the time.
Caps do not do anything to help with congestion issues at the cable node and that's why even with them in place congestion still happens.
Since then they just use speedboost, giving you extra bandwidth for webs surfing and short downloads if the node is not congested, instead of giving you a general bandwidth increase.
Now the caps are left in place purely to limit video competition against cable tv.
Re: It is sad they resisted this change Why are ISPs rolling out broadband if they don't want people using it? at least Google has clarified the TOS to make it more friendly for everyday consumer use. Most other ISPs have not and will not do so.
Re: It is sad they resisted this change This has nothing to do with a commercial for profit ISP using Google Fiber.
Re: Residential vs. Commercial?!?
said by Bengie25:However, an ISP can surely isolate residential service to a prefix that is not announced to certain BGP peers. Meaning they can provide higher quality routing to their business customers and leave the cheap (often congested) peers to the residential subs.
"3) Higher packet queuing priority at congested links.
4) Ability to get a static IP or even the assignment of an entire subnet.
5) No blocked ports.
6) Lower over-subscription ratio."
Even these don't affect pricing that much. I get all of this on my residential line and I pay little more than I did for cable.
Well, #3 is moot. Your ISP can't QoS outside of their network and if the ISP doesn't have congested links, then QoS doesn't do anything.
80/20 rule is that QoS is more expensive than bandwidth for high speed links.
Not all ISPs do this, but I'd imagine it's quite common with the large companies.
Wisconsin Rapids, WI
Re: Residential vs. Commercial?!? I fully agree, but giving my anecdotal situation, not to mention my ISP is quite small compared to incumbents.
I'm in a small city quite far from any metro area. The nearest exchange is Chicago, a state away. My ISP doesn't really do any peering, except with another smaller ISP, but I'm not sure the business relation.
Nearly every route is over the same route, strait to Level 3 in Chicago. I have yet to experience any congestion ever, except in a situation or two that seemed like a peering link failed(massive packet-loss) far-far upstream with L3, but was transient and was fixed in less than 1 hour all of those few times.
My ISP actually advertises "dedicated bandwidth" for all of their line services. So regardless of different routes, they indirectly advertise congestion free Internet access, within their network and to their peers.
My Internet connection is more expensive, but not by much. Local cable advertises $30, with certain restrictions like bundling etc, but just comparing directly, cable is about $30-$60 for 30mb/4mb and my current ISP is $70 for 30mb/30mb.
They actually don't differentiate between business and residential. We all use the same connection and have the same features. All symmetrical dedicated bandwidth with access to static IP addresses /29 for $10/m.
According to my 24/7 ping, I am getting 0.08% packet-loss to Chicago, and less than 0.1ms jitter, 24/7. I have to say it's dedicated. This is including running BitTorrent upwards of 40mb/s on my 50mb connection. Much more than 40mb, and I start to see ping spikes up to 18ms. The 0 downtime over the past 6 months is quite nice to.
I'm not sure that 0.08% packet-loss properly reflects my connection. I did a 24/7 ping for 5 days from my work to my home, I got 0.1% packet-loss. Once I add in that my work uses Charter, and the route is over 2,000 miles long because Charter's peer was routing through Dallas Texas, with the 4x the distance and hops, 0.1% seem not bad at all.
Even then, I typically go days without any packet-loss from the ping, and it's not packet-loss as in loss per n packets, because I can transfer 45mb/s for hours with 0 packets dropped. I'm not really sure where the loss occurs either. As far as I know, I have absolutely no loss caused by my ISP.
My latest test showed me getting about 0.25ms of jitter to LA at 7pm on a 5,000 mile round-trip. I'm not sure how much better one can get.
If it's a question between paying $60/m and $70/m for the difference between congested shared bandwidth and non-congested dedicated bandwidth, I'll gladly pay the premium.