If you're lucky, your broadband connection gives you 1.5MB per second downstream. Imagine if your connection was 1,600 times as fast -- or 1.5GB per second?
Such fast and furious connections aren't as far off as you might think. With encouragement from initiatives such as the
CENIC Broadband Vision for California, such speeds are expected to be available as early as 2010. In the meantime, the current Internet2 Land Speed Record would allow for the transfer of an entire DVD across the Internet in 18 seconds. That's 200 DVDs per hour.
The Internet2 Land Speed Record (I2-LSR) competition for the highest-bandwidth, end-to-end networks is an open and ongoing contest set up in the year 2000. I2-LSR entries are judged on a combination of amount of bandwidth used and distance covered end-to-end using standard Internet (TCP/IP) protocols. The four categories of awards are single and multiple parallel stream using IPv4, and single and multiple streams using next-generation
IPv6 protocol.
The team
The records
We set the single and multi-stream IPv4 records from a single host in Amsterdam to a single host in Sunnyvale, California, on November 19, 2002. Over a period of almost a minute, we transmitted 6.71GB from Amsterdam to California, a distance of 10,978 kilometers (more than 6,800 miles) at an average transmission rate of 923MB per second.
On February 28, 2003, a similar team consisting of scientists from Caltech,
CERN, SLAC, and
Los Alamos National Laboratory set a new single and multiple stream IPv4 record by transferring over a terabyte of data in just under an hour from Sunnyvale to Geneva, a distance of 10,037 kilometers. This was an average data rate of 2.38GB per second and represented and efficiency of over 99 percent of the 2.5GB per second
OC12 bottleneck on the path.
Supporting cast
In addition to the teams of scientists, there were also loans of equipment, fiber links, and facilities from industry and organizations.
Hardware and software
Both records were accomplished using commercial off-the-shelf hardware and software. The hosts were dual 2.4-MHz cup Intel Xeons running
Red Hat and
Debian/GNU Linux with Network Interface Cards from
Syskonnect and Intel. Cisco and
Juniper provided the routers, and transfers were made using the memory-to-memory TCP transfer applications
iperf and
rapid.
Who benefits from faster speeds?
High Energy and Nuclear Physics (HENP) is currently the driving application for this work. HENP is today copying terabytes of data between collaborators in the United States and Europe using high-speed networks. The data rates are roughly doubling per year. Here are some other data intensive sciences with similar needs.
- Astrophysics
- Bioinformatics
- Global weather prediction
- Fusion
- Seismology
- Aerospace
- Medicine
- Security
The ability to transmit data over networks, as opposed to sending literally truckloads or plane loads of data, will reduce costs by automating processes, reducing storage space, and dramatically reducing shipping costs. This ease of transmission, in turn, will open up new avenues for improved collaborations and data mining. Using medicine as an example, expedient access to shared data will result in greatly improved patient care. In the future, we can see dramatic impacts on the media distribution industry when one is able to transmit full-length DVD movies in a matter of seconds.
What does this mean to you?
The importance of the records is in raising the bar on expectations for applications and users. Further, these records demonstrate that the network is rapidly becoming a catalyst rather than a bottleneck for applications. In fact, the bottlenecks are moving to the host and disk sub-systems. It's also significant to demonstrate that TCP can scale from modem speeds to Gbits.
It is important to recognize that the current records required the efforts of typically 20 experts working around the clock. Even then, we were unable to accomplish all of our goals. A major future objective, besides increasing the bandwidth available and throughput achievable, is to turn these efforts from "hero" experiments into mundane, easy-to-perform activities.
