Slashdot Mirror
German Cable ISP First To Deliver 4700Mbps Internet Connection
Mark.JUK writes "It's enough to make grown IT workers cry. German cable operator Kabel Deutschland claims to have become the first provider to successfully achieve a real-world internet connection speed of 4700Mbps (Megabits per second) after they hooked up to a local school's test account in the city of Schwerin. The ISP, which usually delivers more modest speeds of up to 100Mbps to home subscribers, used its upgraded 862MHz network, channel bonding, and the EuroDocsis 3.0 standard to achieve the stated performance. But don't expect to get this kind of speed tomorrow; right now there's no demand for it among home users, and you probably couldn't afford the bandwidth anyway." ("No demand at its current price," at least.)
They used 12 modems and thus 12 seperate channels which means in reality, they only transmitted about 400mbit per "subscriber" (cable).
While this is nifty, Kabel Deutschland subscribers' bandwith is often shared, which means at peak time you don't even get 30 of the promised 100mbit. In addition to that, they slow you down after a 10gb quota/day. And in addition to that, they often throttle certain protocols, namely torrent.
This is one of the worst ISP in Germany who just made a totally useless world record.
It brings a tear to my eye to see the "modest" and "100Mbit" used in the same sentence. Yes, I realize that compared to 4700Mbit it is but I just got upgraded to 5Mbit so I still think you're insensitive clods!
And yeah, I'm sure I could find a use for 4700Mbit.
I have a 200Mbps connection at home, and for now it's fine for, well, everything. But it's really hard to tell what kind of speeds will be useful in the future. Let's imagine a virtual tailor service... Assume that you could go online, image yourself with a high res 3D webcam, and order custom clothes, complete with a virtual mirror to try them on. I'm guessing my 200Mbps connection would fall short at that point.
This is the kind of chicken and egg problem we have with broadband right now. The next generation of online services can't be profitable because the infrastructure isn't there, and at the same time there is "no demand" for really high speed connections, since there are no services that need them.
But at least one can stay positive and hope for cool sci-fi tech, right? ;)
.: Max Romantschuk
What's the point if all they are going to do is cap your usage.
I'm god, but it's a bit of a drag really...
Never underestimate the bandwidth of a truck full of backup tapes.
4700 Mbit/s = 4.7 Gbit/s, how's that a record? The Gathering here in Norway had a 200 Gbit/s Internet Connection, topping Dreamhack in Sweden's 120 Gbit/s. Maybe it's some silly 4.7 Gbit over cable, but that's like the wold's fastest subcompact. And for all of us that have fiber to the home, yeah we know it's just what equipment you put on both ends. The cable itself could probably pull 100 Gbit/s with the right equipment.
Live today, because you never know what tomorrow brings
He fixes the cable?
I hear the latency is pretty high, though.
Finally had enough. Come see us over at https://soylentnews.org/
Prices here in Norway:
Uncapped 5 Mbps ADSL: Around $50
Uncapped 60/60 fiber: Around $90
Okay quite a bit more than you're paying but Norway is in general an extremely expensive country overall, an average full time salary is $75k so by our standards it's cheap. And I once downloaded a 500GB torrent, it really is uncapped. And this country has a population density of 13/km^2 as opposed to India with 368/km^2, delivering broadband there should be much much cheaper. I honestly wouldn't worry it seems mostly like a US problem, all of Europe is constantly upgrading. For example here's from an article I recently read on Britain:
BT said that 7 million premises are now on its fibre network, and this year that number will grow to 10 million. The ultimate target is two-thirds of the UK by the end of 2014.
Oh and they'll also triple top speed from 100 Mbps to 300 Mbps. Any new apartment block or any new housing field is wired with fiber and it's being retrofitted to a lot of old housing too. It's not a question of whether it's the future, but how long it'll take.
Live today, because you never know what tomorrow brings
Yup, we Swedes will always be strong in the "internet scene." A friend of mine just got his 1GBPS connection :( And he pays half my price :
I will tell you from my time working in a company that designed Ethernet PHYs and MACs, that most high end desktops and consumer gear can only maintain a 1Gpbs link, but can accommodate no where near that much BW. The best PCs can only sustain ~500Mbps throughput. Most on-board LAN and sub $100 PCIe LAN cards fall closer to 200Mbps. This is because they do not support DMA and are using Polled IO and the host OS for the LAN stack, much like the old winmodems did.
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
Translation: Since I will lose the argument vis-a-vis USA vs. The Way Everyone Else Provides Internet, I will shut down the conversation preemptively by spouting something meaningless yet somehow jingoistic. Good for you.
PCIe x1 has the same throughput as PCI32/33, so you're going to see the same issues. I have TWO NICs in my system (three if you count the LOM).
The PCIe x1 NIC gives me 500Mbps, the PCIe x4 one with TCP offload gives me 950Mbps on the same workload (server they are connected to has a TCP Offload NIC as well.
The LOM NIC uses a single PCIe lane, so again it's slow.
Sorry to disagree with you, but with a IEEE compliant 1518 byte frame on a normal NIC you're going to have crap performance. Yes you can use jumbo frames if your app supports it and you will see vast improvements, but a TCP offload engine on the NIC will do even better.
-nB
whois gawk date unzip strip find touch finger mount join nice man top fsck grep eject more yes exit umount sleep dump
Well, almost all of the things your listed were built by our grandparents. Since our grandparents didn't build fiber-to-the-house infrastructure, apparently it is not possible for us to even attempt to do so.
And a lot of those things, such as roads, garbage service, water supply, rail, and power, were built by our governments using tax money - which is absolutely forbidden in today's climate of economic religious fundamentalism which demands that all infrastructure creation and related services must be done by entrepreneurs - who have shown they cannot do what the governments of our grandparents did for less than an infinitely growing multiple of what the grandparents paid. Governments build and maintain for the lowest possible cost for the maximum possible return, while our new privatization model demands lowest tolerable service levels for a maximum, and ever-growing per quarter, return on investment. We will never have fiber to the house - with the exception of the very wealthy, of course.
I will tell you from my time working in a company that designed Ethernet PHYs and MACs, that most high end desktops and consumer gear can only maintain a 1Gpbs link, but can accommodate no where near that much BW. The best PCs can only sustain ~500Mbps throughput. Most on-board LAN and sub $100 PCIe LAN cards fall closer to 200Mbps. This is because they do not support DMA and are using Polled IO and the host OS for the LAN stack, much like the old winmodems did.
-nB
I find your statement about lack of DMA and low throughput very hard to believe. I worked at Marvell Germany until September 2007 as a device driver developer and *all* of the then-current Marvell Gig chips (Yukon-II) *easily* managed 900+ MBits/sec, the Windows driver actually peaking at 980 megabits/sec. This was both for onboard controllers (e.g. Asus Mobos) as well as Ethernet cards, for all operating systems supported (Windows*, Linux, Mac OS X, FreeBSD, Solaris, Aix and HP/UX).
Furthermore, *all* the drivers used DMA; only the Linux and Windows drivers offered the option of polled operation to *increase* throughput (no, not a typo, although it's counter-intuitive).
Note, however, the above numbers apply to proper Ethernet speed testing, using an in-memory data generator for the transmit side and a corresponding receive program on the receive-side, avoiding any reads or writes to a disk. Otherwise you'd just be testing the disk read and write speeds, which was a common mistake made by testers, and could well account for your low cited speeds.
Meaningful throughput tests may be performed by tools such as ttcp.
--Gerald