Slashdot Mirror
Researchers Break Internet Speed Records
MosiMosi wrote to let us know about a new development on the Internet2 front. Researchers in Tokyo have advanced the speed of the network, breaking records twice in two days back in December of last year. "On Dec. 30 [researchers] sent data at 7.67 gigabits per second, using standard communications protocols. The next day, using modified protocols, the team broke the record again by sending data over the same 20,000-mile path at 9.08 Gbps. That likely represents the current network's final record because rules require a 10 percent improvement for recognition, a percentage that would bring the next record right at the Internet2's current theoretical limit of 10 Gbps."
9.08 * 1.1 = 9.988
...the future crusty old bastards are already drinking the Kool-Aid.
Let's see... A DVD has 3.76e10 bits. If you drove 20,000 miles at 70mph, that would take 1.03e6 seconds. So each DVD in the wagon would give you about 37Kb/s bandwidth. So you'd need about 248,000 DVDs in the car. My little postal scale here says that a DVD in a paper sleeve weighs in at 20g, so you'd need almost 5000kg of DVDs, which is probably too much for a station wagon. You could probably manage the task with BluRay or HD-DVD, though.
If you are going to use throughput it probably gets a bit scary; raw bandwidth, sure, but I can click send faster than you can burn your blu-ray discs. Your throughput is limited to some multiple of the fastest burner money can buy. If you magic 10x blu-ray burners into existence, you would only need a dozen or so to keep up with the network(of course, you have to go way, way faster than the network because you have all that dead time while you bring the discs over to your readers at the other end of the transmission). And they did it with a good installed circuit, which may or may not represent the latest advances in network architecture.
Here's a good theorem though: People care a lot more about latency than bandwidth.
Nerd rage is the funniest rage.
As I understand it, this is over one link. OC-192 is actually a series of OC-48 links bonded together.
Heck you can get yourself a nice 10gbit/sec line with 10 1gbit lines, ooh la lah
Tom
Someday, I'll have a real sig.
You are mixing up latency with bandwidth. The latency (round trip time) of the connection here is 400 hours. The bandwidth (i.e. data rate) is the amount of data divided by the time it takes for the data to travel its own length.
At 100 mph, a station wagon will travel its length in 0.14 seconds. So the bandwidth of a stationwagon packed with 9000 GB of data is about 550 Tbps.
Given a train of station wagons running at 100mph, you could sustain that. Of course with 1440000000 ms ping times, I wouldn't try playing Battlefield 2 over that connection.
Seriously, the distinction is important. If you included transit time when calculating bandwidth, the theoretical maximum bandwidth for a 12,000 bit packet on a 20,000 mile path would be 112 kbps.
Support SETI@home
Ummm, OC-192 is 9.6Gbps I think they are a little shy of the speed record. Maybe I missed something.
Within a data center or a metro area, it's commercially viable to pump tens of gigabits per second of bits from point A to point B using many parallel fiber circuits between the two locations. What makes the Internet2 land speed record (http://www.internet2.edu/lsr/) interesting is adding distance to the problem by multiplying the speed times the distance. The unit of measurement they use is "terabit-meters per second" (Tbmps?). The current record is 272,400 Tbmps, or ~9Gbps over 30000km (1km=1000000m). The transfer rate is really a function of 1) latency adjustments in the data transfer protocol, 2) the minimum transfer speed capable between all points on the network (currently OC192=10Gbps), and 3) the speed of the sending and receiving computers. While OC192 might theoretically be 9.6Gbps, getting the various vendors
switches on different continents to all send packets at line speed for a long period of time with minimal packet overhead can be challenging.
What makes this pointless, though, is that the sending and receiving equipment is in the same location. In their documentation they send the bits from a computer in Tokyo through Chicago through Amsterdam and back through Seattle to the same lab in Tokyo. It would be much easier to put a 10GigE fiber between the two machines, but that's not he "point" of the exercise.
Someone has to pay for this. Usually its the country's taxpayers or a company's stockholders.
I'd much rather see benchmarks for transferring N terabytes of real data from one site with lots of disks to another far-away site with lots of disks. Real companies can use that data for pontificating disaster recovery and content/database replication technologies. I'd reckon that Google can beat the multiple stream Internet2 LSR any day they want by pumping petabytes of data between its data centers over multiple 10GigE backbones. Andy Tanenbaum's (or Hal Stern's?) station wagon full of tapes is also a fine competitor.
-ez
You look at a progress bar for entertainment?
On a more serious note, you'd still get in one chunk, so the initial byte wouldn't matter.
What we need to look at is Gigabits per dollar.
Assuming that you were somehow blessed with an ISP that would let you download over a TB/month, and had a 5Mb/s connection (and assuming constant speed), it would take roughly 19.4 days to download.
Assuming a 30-day month and that your ISP charged $40/month, it would come to $25.86 for that one transfer, which would be $0.000003/Mb.
Shipping a 700g package to (anywhere in) Canada via USPS airmail (The Internet is international, after all.) would be $14.50. That comes to $0.000001/MB.
Just my 0.0002 cents.
I just read Slashdot for the articles.
Why is this "10G" even news? 10 Gigabit (OC192) Has been around since at least 1999. In fact, engineers & scientists already have functioning proto-types of 100 Gigabit over fiber (basically DWDM - multiple colours of 10 Gigabit streams multiplexed).
l y/art.php?2642
The IEEE expects the standard to be ratified in mid 2008 for the fiber version & copper (CAT8?) to come out within a couple of years after that (late 2009 or 2010).
Siemens achieves 111 Gigabits over 2,400 kilometers
http://presszoom.com/story_127837.html
Bell & Lucent labs acheive 107 Gigabits over 2,000 kilometers
http://www.enterprisenetworksandservers.com/month
Wikipedia:
http://en.wikipedia.org/wiki/100_gigabit_Ethernet
Those Internet2 people are just a tad behind... like 10 fold! If Internet2 = 10G, and Internet3 =100G, then really those Internet2 people should be working on Internet4 (Terabit baby!)!
Adeptus
No trees were killed in the making of this post; however, many trillions of electrons were horribly inconvenienced.