New technology enables high-speed data transfer
June 18, 2009GridFTP, a protocol developed by researchers at Argonne National Laboratory, has been used to transfer unprecedented amounts of data over the Department of Energy's (DOE) Energy Sciences Network (ESnet), which provides a reliable, high-performance communications infrastructure to facilitate large-scale, collaborative science endeavors.
The Argonne-developed system proved key to enabling research groups at Oak Ridge National Laboratory in Tennessee and the National Energy Research Scientific Computing Center in California to move large data sets between the facilities at a rate of 200 megabytes per second.
The deployment of GridFTP at the two computing facilities is part of a major project to optimize wide-area network data transfers between sites hosting DOE leadership-class computers.
According to Ian Foster, co-director of the Globus Alliance project responsible for designing GridFTP, large-scale data transfer places an enormous burden on networks. "Conventional protocols have proven unable to handle the increasing demand of large-scale data transfer," he said. "The result has been delays in obtaining data, or even lost data as the network becomes overwhelmed. GridFTP changes that."
As large-scale collaborative science projects become increasingly common, the need to transfer unprecedented amounts of data is becoming critical. Having GridFTP on ESnet will enable the sharing of data between supercomputer centers in disciplines such as climate modeling and nuclear physics that require secure, robust, high-speed bulk data transfer.
"Our goal is to enable the scientists to rapidly move large-scale data sets between supercomputer centers as dictated by the needs of the science," said Eli Dart, a network engineer for ESnet, which is managed by Lawrence Berkeley National Laboratory. "High-performance networking has become critical to science due to the size of the data sets and the wide scope of collaboration characteristic of today's large science projects such as climate research and high energy physics."
GridFTP offers several advantages over other data transfer systems. For example, with Secure Copy, or scp, bulk transfer of a 33-gigabyte dataset between the two remote hosts could take up to eight hours. With GridFTP, almost 20 times that amount of data can be transferred in the same amount of time. And, unlike the transfer application FTP, GridFTP uses multiple data channels for improving the transfer speed.
"The data tsunami problem has been a major bottleneck to scientific advancement," said Raj Kettimuthu, technical lead and technology coordinator of the GridFTP project at Argonne. "With GridFTP computational scientists can analyze their simulated and derived data in real time."
More information on GridFTP is available at http://www.globus.org/grid_software/data/gridftp.php .
Source: Argonne National Laboratory
Sep 01, 2005 |
not rated yet |
0
There are times when it would be useful for the average end use to be able to multi-stream from a single site. A lot of sites have a per connection speed cap to keep one person from hogging the bandwidth.
The real question is when will someone build a an online data cache service. Cache parts of files on servers all over the country so when a subscriber goes to download one, they get the parts multi streamed from all over. Like an on demand bit-torrent for popular data using dedicated servers. It's just a question of the business model working with bit-torrent as a free competitor.
There was an article about 5 months ago that Microsoft was funding a company in or near Fresno,CA that was doing this very thing you mentioned.
I don't have the link anymore but should be easy to find since the Central Valley have very few technology companies there.
Most of the hosts you speak of will limit by IP, not by each connection. So, this technology still wouldn't really help.
This is pretty much a CDN in reverse, and I bet you could create an ad-hoc system using a current CDN to do just this where instead the user uploads it's file onto the CDN and then the server is told to download the file from the CDN.
This would still not avoid bandwidth caps on the destination server.
The company I spoke of from Fresno is doing something different from the ones you spoke of. Microsoft thought it was unique so much that they funded some millions to assist with the equipment.
Wish I can find that link.. arrrr