Slashdot Mirror
Will Fiber-To-the-Home Create a New Digital Divide?
First time accepted submitter dkatana writes Having some type of fiber or high-speed cable connectivity is normal for many of us, but in most developing countries of the world and many areas of Europe, the US, and other developed countries, access to "super-fast" broadband networks is still a dream. This is creating another "digital divide." Not having the virtually unlimited bandwidth of all-fiber networks means that, for these populations, many activities are simply not possible. For example, broadband provided over all-fiber networks brings education, healthcare, and other social goods into the home through immersive, innovative applications and services that are impossible without it. Alternatives to fiber, such as cable (DOCSYS 3.0), are not enough, and they could be more expensive in the long run. The maximum speed a DOCSYS modem can achieve is 171/122 Mbit/s (using four channels), just a fraction the 273 Gbit/s (per channel) already reached on fiber.
google can advertise 1gbps all they want but the truth is that if you're poor, a 5-10mbps connection is more than enough to get your kids watching educational shows on netflix and youtube. and torrentfreak had an article today about google drive and dropbox and onedrive throttling people on their end because most end points cannot support their user base at 1gbps per user. all the servers are virtualized and over subscribed in the cloud, they aren't built for performance.
i know people on 10mbps connections and their netflix comes in perfect HD because their ISP has an open connect appliance or a direct connection. staging the content close to the user and not streaming it via level 3 will do more for performance than paying for 100mbps or some ridiculous speed just to save 10 minutes on your steam downloads
it's like the PC market in 2001. the P4 is more than fast enough for anyone at the time
TCP_WINDOWS_SIZE can grow up to 2GB which is enough for a 10GB link with 1600ms latency. I'm not going to say that your OS will be happy about it, but that's the logical limit. As for AES 256 encryption, a modern desktop CPU can handle 100mb-300mb per core, unless you have AES-NI, then it's more like 1gb per core. OR if you're like me, you NIC supports line rate IPSEC offloading, so 4gb/s with 0% cpu overhead, assuming IPSEC and not VPN.
Not having the virtually unlimited bandwidth of all-fiber networks means that, for these populations, many activities are simply not possible. For example, broadband provided over all-fiber networks brings education, healthcare, and other social goods into the home through immersive, innovative applications and services that are impossible without it.
I think this point requires further explaining.
Why exactly do I need Gbit service to bring healthcare into my home?
Alternatives to fiber, such as cable (DOCSYS 3.0), are not enough, and they could be more expensive in the long run. The maximum speed a DOCSYS modem can achieve is 171/122 Mbit/s (using four channels), just a fraction the 273 Gbit/s (per channel) already reached on fiber.
Huh?
DOCSIS 3.0 does not have a maximum limit on the number of channels that can be bonded.
The initial hardware would only bond up to 8 channels (~304 Mbit/s), but 16 channel (608 Mbit/s) hardware is already being rolled out by Comcast in the form of rebadged Cisco DPC3939 Gateways.
2015/2016 we might see 24 channel (912 Mbit/s) and 32 channel (1.2 Gbit/s) hardware.
2016/2017 is most likely, in the form of DOCSIS 3.1 modems, which use completely different modulation, but will have 24/32 channel DOCSIS 3.0 baked into them so that the ISPs can seamlessly upgrade from DOCSIS 3.0 to 3.1.
Cable's game plan is to use DOCSIS 3.1 to put off pulling fiber to the home, which keeps their costs low and will allow them to offer (multi)gigabit speeds using a hybrid fiber/co-ax infrastructure.
[Fuck Beta]
o0t!
1) TCP alternatives are already being developed
2) TCP_WINDOW_SIZE problem was solved long long ago with TCP_WINDOW_SCALING. The limit is roughly 100 Gbit/s at 80ms
3) Not sure where you're getting your data from but reality is a very different place from where you live. 25 MBps would be an Intel Atom 230 decrypting AES-128-CBC. 5 year old mobile/low power processors were never meant to stand the test of time. Take something from around the same period, like say an Intel T5550 and all the sudden you're up to 80MBps for AES-256-CBC (or 109MBps AES-128-CBC). Even dropping down to a P4 you can get 75MBps for AES-256-CBC.
Also, arc_four (aka RC4) is not even worth discussing as it's completely useless as encryption. RC6 is (comparatively) fast at low byte counts on specific platforms but quickly plateau with little performance increase after 128 bytes and slows by a factor of 3 if the hardware is not optimal. Rijndael, which was chosen for AES, had consistently fast speeds no matter the bytes or platform. The reality is that any chip with the AES-NI instruction set makes it a moot point, by example the i7-3960X is churning out 5.7GBps. Without it, performance does suffer, but you're still talking 250-400Mbps on a 4 core chip.
The real question is how the heck this got posted to Slashdot. DOCSIS 3.1 bumps the limits to 10 Gbit/s down, 1 Gbit/s up and even on DOCSIS 3.0 - who says you've got to be stuck at 4 channels? 24 Channel is already actively deployed in Canada at 200-250Mbps down/15-30Mbps up, 1.5Gbps/150Mbps in the UK.
Looks like it can go higher. A quick Google search found TCP window scale option, which apparently has been supported by all the major OSes for a while (since Linux 2.6.8 and Windows 2000).
A single fiber cannot handle the world's internet bandwidth
Current state of the art is 1pb/s over a single fiber, about 10x the speed of the Internet. Obviously impractical for a single fiber to connect every house in the world. The Internet is about 100tb/s right now, you can get 30tb/s over a single fiber with commercially available technology.. So 3 fibers?
And the PON systems used for homes don't even dedicate 1Gbit to each termination
WDM-PON, which is what Google Fiber uses, is 40gb/40gb with 32 lambdas of 1.25gb/1.25gb each, given each end point it's own 1.25gb/s.
You don't have a dedicated connection to a chassis with 2,000 other customers, you are PON split from a single fiber with a lot of other houses
With GPON this is the most common setup, but WDM-PON is backwards compatible with regular GPON. The most common setup is dedicated fiber back to the CO, which means an upgrade to WDM-PON is as simple as switch out the line card, then placing lambda filters on each customer's fiber, which is done back at the CO. I've called up my ISP and had them change which GPON port I was plugged into, took them about 5 minutes from the time the tech said "give me a second".
I think it's hilarious that you think that your ISP is only oversubscribing their links 2x (2,000 1Gb connections to 1Tb backhaul). That's fantasyland at the prices that residential customers pay.
I wasn't talking about the backhaul, there is no backhaul. Fiber is best described as a "Non blocking consolidator that plugs directly into the trunk". My ISP has a 3x undersubscription, in that the trunk is 3x the peak monthly peak.
Fiber can't fix bad designs, but fiber lends itself naturally to cheap, easy, and scalable designs. A single consolidator/chassis can support 2,000+ customers with 3tb+ of bandwidth. If the ISP only uses 1gb uplinks, then they're screwed anyway. But for a one time cost of $6k, you can purchase a 100gb port. They're not expensive anymore. The point is fiber makes it retardedly simple to have the entire bottle-neck be the backbone, instead of some complicated mixture of middle-mile nodes and shuffling around customers.