New Data Transmission Speed Record

An anonymous reader writes "Gizmag is reporting that a team of German and Japanese scientists have collaborated to shatter the world record for data transmission speed. From the article: "By transmitting a data signal at 2.56 terabits per second over a 160-kilometer link (equivalent to 2,560,000,000,000 bits per second or the contents of 60 DVDs) the researchers bettered the old record of 1.28 terabits per second held by a Japanese group. By comparison, the fastest high-speed links currently carry data at a maximum 40 Gbit/s, or around 50 times slower."

Why is bandwidth measured in Kb

[b00m3rang]

and storage in KB? It's not like we measure radio waves in cycles per second and sound waves in cycles per 8 seconds. What's the advantage?

Filesystem and Ultra320 SCSI are our chokepoint

[PSaltyDS]

We have a 30TB EMC CX-500 with Brocade 2Gb FC backbone. The bench moves blocks of a few hundred GB to a dozen servers or less. We have never come anywhere near %50 utilization on the FC.

The transfers run about 4-6hrs and I was looking for choke points to shorten the time. The data simply won't go to disk any faster on the U320 SCSI bus. We consistently measure 20MBps max to disk, which makes sense. U320 means 320Mbps/8 = 20MBps. So I get the same max numbers for local disk-to-disk that I get for SAN-to-disk, and the same results regardless of OS. If this rate could be maintained, six servers doing the transfer should just about saturate the backbone, but the overhead of file access and FS management mean the max is only maintained for a moment as a few particularly large files come across. With lots of smaller files being copied, the average rate goes down to 2MBps.

If these servers had to be optimized for SAN-to-Disk transfer rate, they would have to have multiple SCSI controllers and HBAs, paired up on seperate PCI busses, and the data would have to be optimized with fewer/larger files.

Of course, the 2.5TBps link is of interest to ISPs and regional carriers not server labs, but I thought I'd throw in what we've seen on the utilization of a 2Gbps FC link in a SAN setup.

Re:Try 10,000 Kilometers

[DRUNK_BEAR]

Optical fiber gives you a loss of approx a quarter dB per km (0.25 dB/km) - which is very close to the theoretical limit of current glass optical fibers. At 160 km, that gives you an attenuation in your optical signal of about 40 dB. All-optical amplifiers - EDFAs - (without signal regeneration, just plain amplification) can give you a good boost in power and you can cascade many before the signal becomes too distorted (because each amplifier amplifies both the noise and the signal). An other type of all-optical amplifier - DRAs - give you a lesser gain, but they do not amplify the noise (pretty cool concept!), so a combination of EDFAs and DRAs can get you an all-optical link - without regenerators - of several thousands of kms! Research scientists do work in a modular fashion. One research group works on one aspect of a problem and other groups work on other aspects. In this case, if one span of optical fiber (without amplifiers) can handle very high bit rates, it is not too difficult to extend this to ultra long haul (UHL) networks which have many spans of optical fibers with all-optical amplifiers - provided that you have amplifiers available for the band of wavelengths that you are using. I am guessing that this is a limitation to this experiment, but a lot of research is done by other groups on all-optical amplifiers working in different wavelength bands. One group cannot do it all, but they rather concentrate on one area and let other groups develop the other areas. So this is indeed a pretty cool and impressive result if we understand the experiment and can see its implication for future optical networks!