Slashdot Mirror
Where's Our Terabit Ethernet?
carusoj writes "Five years ago, we were talking about using Terabit Ethernet in 2008. Those plans have been pushed back a bit, but Ethernet inventor Bob Metcalfe this week is starting to throw around a new date for Terabit Ethernet: 2015. He's also suggesting that this be done in a non-standard way, at least at first, saying it's an opportunity to "break loose from the stranglehold of standards and move into some fun new technologies.""
I'd like to see the internet held together by his fun new technologies. See how well machines communicate without basic protocols.
640k is roughly:
10 Commodore 64s
20 BBC Micros
640 ZX-81s
6 times a SDSS floppy disc
Who needs that kind of memory?
We might not need terabit ethernet *now*, but in 25 years time, it may be the basic expectation of your LAN's speed.
Oolite: Elite-like game. For Mac, Linux and Windows
.. a technology that lets homes receive fast internet no matter where they are. My area's not cabled up, and thanks to me being too far from the exchange.. I just live in a normal street .. I can't reliably get more than 512KB a second. Fix that, and you'd be laughing. Powerline networking, maybe?
Every ISP in the world, to meet the bandwidth allocations they've sold fraudulently.
Finally had enough. Come see us over at https://soylentnews.org/
Switches and machines that aggregate multiple saturated gigabit connections?
7 years is a long time.
Seven years is the blink of an eye, kid.
mcgrew's razor: Never attribute to stupidity that which can be explained by greedy self-interest
Regional ISPs. This is not a consumer product. Running ethernet on the backbone allows a homogeneous stack on all hosts from end to end.
640k is roughly:
You miss the point.
There is nothing you can do with a big-ass pipe except move bits.
Plugging your firehose into the neighborhood drip irrigation system isn't going to get your lawn watered any faster. In situations where insane bandwidth can be installed end to end and there are insane amounts of data to move, this would be a great thing. However, the GP's point was that this really isn't the most common situation.
Most LANs have TONS of bandwidth to spare today. Work on an Internet (both pipes and servers) that can keep up with my cheap commodity HW LAN. Now, THAT would be useful.
This is not to say that there is no one on earth that needs to move insane amount of data in a LAN. Good on them. However just that increasing LAN speed won't help most folks while they wait for "the network"
Maybe. One of the things that I've noticed is that as the bandwidth increases it becomes harder and harder to fill it up. Back in the Commodore 64 days it was not hard at all to run your machine out of memory by just typing a paper that was too long, and that's without graphics/charts/etc... These days there is no way a person would be able to type enough text to even make a noticeable dent in the main memory of any commodity machine. When everybody used 56k modems and serial lines it was trivially easy to fill up the link. However, when they moved to 10Mb Ethernet it got harder, but not impossible. Suddenly compressed music files were not a problem, although compressed video (DivX) still was. Then we went to 100Mb Ethernet and compressed video is no longer much of a bottleneck. Even now most modern machines come with Gigabit Ethernet ports that your average person can't fill with anything. Without new and bandwith intensive applications people won't be inclined to improve their bandwidth.
That's not to say someone won't come up with some application that requires a ton of bandwidth (distributed neural nets?), but none of our current applications would even really scale up to requiring 10GbE. The only realistic thing that comes to mind is some sort of Super HD video format, but anything like that is at least a decade away.
I read the internet for the articles.
1Gbs is a bit slow when backup up a 1TB hard drive to the network server at home. ;-)
I humbly submit that the R&D money that could have increased the upper boundary of Ethernet speeds was spent to bring wireless to the masses. Five years ago, if you'd told me WiFi would now be a year away from nominal speeds of 250Mb/s I might have thought you were talking about prototypes. The dorms where I was a tech had just finished upgrading from 10Mb/s to 100Mb/s Ethernet. The few laptops that were sold with external wireless cards had nominal speeds of 10Mb/s. But now we have 802.11g and next year we should have 802.11n on the store shelves.
I think we've gained much more by pushing out the median speed of wireless than we could have gained from pushing out the marginal speed of twisted pair.
One might ask the same about birds. What ARE birds? We just don't know.
It's not a question of your end points needing that kind of bandwidth, it's a question of the links between ISPs and such. Look at any of the large router manufacturers today and see what kind of interfaces they're putting on their high-end gear: multiple 10Gbps ports. You can safely believe that if 100Gbps were available now that people would be using it. The step to 1Tbps is a large one, but there's no such things as "too much" bandwidth.
I read it on the Internet, it must be true!
There is nothing you can do with a big-ass pipe except move bits.
Free clue: 10Gb ethernet is currently used mostly in clusters and as backbones for large network installations to move lots of data around very fast. It's a long way off being a LAN technology. In seven years time, Terabit ethernet will be used mostly in clusters and as backbones for large network installations and 10Gb ethernet will be a LAN technology.
- over 10 DEC PDP-11/45s running the RSTS time-sharing system
The maximum memory on these things was 28K words (16-bit) without memory extension hardware. In the 70s we had 8 users on a system with 28K of memory sorting lists, printing reports, data entry, editing with TECO, batch runs in the background at low priority, with relatively few swap thrashing problems. I implemented an ultra-low priority batch mode that waited until there was nothing else running for 5 minutes before activating a u.l.p job, then swapped it out instantly as soon as someone pressed a key - amazing what you could do with its sys calls. I did all sorts of hobby computing in this mode (with employer's permission) - looking for Fermat number factors, analyzing stock market timing data for patterns, you name it - I was like a kid in a candy store.
In retrospect, it was simply amazing how they shrunk it into that amount of memory and amazing how much I could do with it. 640K was simply unimaginable at that time.
If you use just one hard drive, that might be true. These days, though, most people with those sorts of storage transfer issues have RAID-5 sources.
I know that the network is my bottleneck when copying from a 7-drive RAID-5 (6x SATA-300 effective data throughput...easily 200MB/sec) through a PCIe 8x host adapter (2000MB/sec) to the gigabit ethernet (125MB/sec).
On the other hand, if I'm not backing up multi-gigabytes of data, normal access over the network isn't much slower than local drives for real-world performance, especially with the large caches in the servers that avoid the disk altogether.