8.6 GB Internet? 382
prostoalex writes "Caltech computer scientists announced the protocol, capable of delivering 8,609 Mbps over the Internet, using 10 simultaneous flows of data. The research project was conducted in partnership with CERN, DataTAG, StarLight, Cisco, and Level 3. The practical applications, according to the press release, is ability 'to download a full-length DVD movie in less than five seconds'. There is a number of papers and scientific publications available."
Nothing new (Score:4, Informative)
Its just that optical routing is expensive and so would the switch at such a high speed.
Re:Argh! 8Gb (Score:5, Informative)
Saying 8.6GB is off by an order of magnitude.
Sigh..
Bloody repeats (Score:4, Informative)
Re:Megabits per second (Score:2, Informative)
RAID (Score:2, Informative)
Not only can a high-end storage array handle that sort of throughput, but it can do it without any bugs [bugfreeliving.com].
Re:Argh! 8Gb (Score:5, Informative)
Re:Nothing new (Score:5, Informative)
This isn't talking about delivering raw speed over a point to point connection but delivering a large amount of data over a shared network. It's talking about a protocol rather than a transport medium, which must account for problems such as error, latency, bandwidth, and flow control.
Overhead (Score:5, Informative)
It is just over a gig a second.
Not all of that is data. Some is packet headers. Some is error correction. That's why you can't push 6 KB per second over a v.90 dial-up connection at 48 kbps.
Re:What's wrong with IP? (Score:5, Informative)
Re:Eight giga-whats? (Score:1, Informative)
Spam emails in my Hotmail account run (on average) 7 kiloBYTES, or 56 kilobits. Now if this connection will give you 6,609 megabits, that's 8,609,000 (roughly) kilobits per second. That's 153,732 spam messages per second. Times 86,400 seconds per day, that comes to 743,817,600,000 - 743 BILLION spam e-mails per day. Of course, that would come to almost 5 terabytes (not -bits) per day, so your disks would, as other posters have mentioned, be the lowest common denominator.
Just FYI.
Re:Argh! 8Gb (Score:3, Informative)
Actually when talking about DVDs, the density of each layer cannot change.
I believe you are referring to SSSL (Single-Sided, Single-Layer) DVDs, as each since on a regular red-ray DVD disc can have two layers of data.
I'm not sure if that's the same for the famed blu-ray discs or not, I would appreciate it if someone could enlighten me.
Re:Bottleneck (Score:5, Informative)
A 33 MHz/32 bit PCI bus which is standard in most PCs will transfer 133 MB/sec.
A 66 MHz/32 bit PCI bus which is in quite a few Intel and UNIX servers will transfer 266 MB/sec.
A 66 MHz/64 bit PCI bus which also is quite common in UNIX servers (and becomming in Intel) will transfer 532 MB/sec.
A 133 MHz/64 bit PCI bus which is the current standard for big UNIX servers will transfer, you guessed it, 1 GB/sec.
Mind you that these numbers are pr. PCI bus, some of the lager Intel servers, and most UNIX servers have more than one PCI bus.
Re:Argh! 8Gb (Score:3, Informative)
Look it up: Dictionary.com definition of order of magnitude [reference.com]
Um. . .Actually it's (Score:1, Informative)
Re:Nothing new (Score:5, Informative)
If you read the article, you'll notice that they're talking about a new transport protocol, not a physical layer. Yes, physical layers have been able to run faster than 8 gig for quite a number of years. But if you run a TCP session across such a link, the window sizes, congestion control algorithms, ack timers, and such features keep you from getting anything like the speed of your physical link.
To illustrate with a point from the article, TCP over the 1 Gbps link they were apparently using typically maintained a throughput of about 270 Mbps, only 27% efficient even when that one session had the whole link to itself.
Previously, no one much cared, because you used those multi-gigabit links by multiplexing a whole bunch of different users; each user saw a much slower link. So, the old, slow TCP was adequate. (To a point, of course; people have noticed problems with TCP acks even on DSL links.)
One of the purposes of the "Internet 2" research and other projects was to develop protocols that would allow _single connections_ to run at very high speeds.
The announced protocol, according to the article, managed 95% efficiency and 925 Mbps with a single flow, and 88% efficiency and the 8.6 Gbps of the headline with 10 concurrent sessions over a 10 Gbps link.
So, yes, it is something new.
Re:Argh! 8Gb (Score:3, Informative)
Base 2 is pointless because comms people use SI prefixes properly*. 1 megabit = 1,000,000 bits. Base 10.
*Probably because not all platforms use 8 bit bytes (encoded to 10 bit bytes for transmission). The comms mfr's only care about the rate on the line.
Re:CalTech (Score:5, Informative)
The problem is that the fastest hard drives on the market today are Ultra320 SCSI, which have a throughput of 320MB per sec... or about 2.5Gbps. Even that's theoretical, of course. And few people have an all Ultra320 datacenter.
Just pointing out that the cabling is hardly the bottleneck when you reach that kind of speed, even at the LAN level. I've seen so many people upgrade their switches to gigabit ethernet then scratch their heads wondering why the network is still slow... when the server in the closet hasn't been upgraded in 5 years. Storage will continue to be the bottleneck on the LAN for a long time to come.
Re:Nothing new (Score:5, Informative)
They achieved this by modifying the TCP protocol to be more efficient. It's a huge improvement.
Of course you'd need a big pipe to the net to be able to handle this bandwidth; it will be quite a while before home users could see this kind of speed. But the new protocol sounds like a big win over regular TCP and hopefully it can be rolled out relatively quickly.
Re:What's wrong with IP? (Score:2, Informative)
Unreliable in this context does not mean you can't depend on it, it means that it is an 'open ended' protocol other wise known as 'best effort' which means there is no delivery confirmation at that level of the OSI model. Basically when an IP packet or an ethernet frame goes out on the wire, that device has no idea whether it got where it was supposed to go. Think about regular postal mail vs. a registered letter.
As mentioned, those protocols rely on higher (higher as in the OSI model) level protocols to do error correction/detection and retransmission requests, such as TCP.
-ft
Re:CalTech (Score:5, Informative)
Re:This sounds like what the Pentagon needs (Score:2, Informative)
Re:Megabits per second (Score:1, Informative)
Some older architectures used 7-bit bytes; some had 9-bit bytes.