Skip to Content

Faster File Flow

Startup FastSoft could speed up the transfer of video and voice files over the Internet.
July 20, 2007

Aria, a new product from California-based startup FastSoft, speeds up the transfer of any large file over the Internet, without requiring hardware or software on the receiving end. According to Dan Henderson, FastSoft’s vice president of product and market development, a 700-megabyte movie file that takes 50 minutes to download regionally via a cable-modem connection can be downloaded in 34 minutes if the sender uses Aria. Overseas transfers show a bigger difference, with the same 700-megabyte movie taking nearly eight hours to download from Asia via a cable modem, and about 45 minutes with Aria.

Transmission system: FastSoft’s Aria appliance, shown here, can speed the transfer of large files over the Internet. The box plugs into a sender’s server and allows any number of recipients to download files at a faster rate than conventional methods without first having to install software or hardware.

Darin Harris, chief information officer of the movie-production company the Post Group, says that his company has seen files that normally take a day to download go through in a couple of hours. Harris says that he was attracted to Aria in part because only the sender needs to own hardware, and recipients don’t need to know anything about Aria to get their files faster. All the sender has to do is connect the Aria appliance to his or her server.

Aria, which is based on research from the California Institute of Technology (Caltech), tries to make the most of the available bandwidth. Data usually isn’t transferred at a line’s full capacity. This is because the transmission control protocol (TCP) that governs 90 percent of the traffic flow over the Internet hasn’t kept up with the times.

A 20-year-old algorithm, TCP is designed to make sure that information sent is reliably received, and to watch for congestion in the network. Standard TCP, sometimes called Reno, works by carefully increasing the rate at which it sends information until it detects that a packet of information has been lost. Reno interprets the loss to mean that the network is too congested to handle that rate of flow, and it cuts its transfer rate in half. Lachlan Andrew, a Caltech researcher who studies the algorithms behind Aria, compares this to a driver who accelerates slowly up to the speed limit, then slams on the brakes. When connection speeds were slower, this pattern wasn’t a problem; now that industrial connections are available at speeds of 155 megabits per second, it is. “Reno is really being stretched to its breaking point,” says Steven Low, cofounder of FastSoft and head of the Netlab research group at Caltech.

Caltech researchers designed FastTCP, the algorithm behind Aria, to improve on Reno. In academic trials, Henderson says, FastTCP has set data-transfer records, transferring data at a sustained rate of 101 gigabits per second. That’s equivalent to transmitting the contents of the Library of Congress in 15 minutes.

FastTCP works by detecting congestion a different way. Reno detects congestion by watching for lost data. FastTCP monitors the real-time transfer speed of each packet of information, watching for delays. This way, adjustments can be made smoothly. Andrew says that the difference is particularly important for wireless networks. Since transfer over a wireless network is often less reliable, packets of information are lost fairly often. FastTCP, he says, detects that there’s no need to slow down the information flow: it simply resends the lost packets.

“In order to be smooth, which is what FastTCP does, you have to be more gentle and assess queuing delay,” says Aleksandar Kuzmanovic, an associate professor in the department of electrical engineering and computer science at Northwestern University who researches TCP. By keeping the transfer from reaching the point of packet loss, he says, it’s possible to avoid sending bursts of data, which can create a jittery effect for the recipient.

The problem, Kuzmanovic says, is that FastTCP isn’t the only attempt to improve on Reno. Other protocols have been designed that could speed file transfer while still using packet loss to monitor congestion. If those protocols send over the same line as FastTCP, they could hog the bandwidth. A protocol that senses delays could be at a disadvantage, Kuzmanovic says, since it might politely back away from congestion where more-aggressive protocols push to the point of packet loss.

Low says that his research group has been studying the issue of fairness–the question of how polite and aggressive protocols should interact. In theory, he says, it is always possible to create a fair situation, even between different types of protocols. Low’s group has been working on an algorithm that could be used to manage how bandwidth is shared among FastTCP flows and other protocols. Although there have not yet been any problems with fairness in practice, Low says, the algorithm could be made part of Aria if the need arose.

On July 30, FastSoft will announce its next commercial product in the Aria line: the Aria 2000, which the company says can accelerate throughput to up to one gigabit per second. Pricing for Aria appliances starts at $10,000 and goes up to more than $200,000, depending on model and configuration.

Keep Reading

Most Popular

Large language models can do jaw-dropping things. But nobody knows exactly why.

And that's a problem. Figuring it out is one of the biggest scientific puzzles of our time and a crucial step towards controlling more powerful future models.

The problem with plug-in hybrids? Their drivers.

Plug-in hybrids are often sold as a transition to EVs, but new data from Europe shows we’re still underestimating the emissions they produce.

How scientists traced a mysterious covid case back to six toilets

When wastewater surveillance turns into a hunt for a single infected individual, the ethics get tricky.

Google DeepMind’s new generative model makes Super Mario–like games from scratch

Genie learns how to control games by watching hours and hours of video. It could help train next-gen robots too.

Stay connected

Illustration by Rose Wong

Get the latest updates from
MIT Technology Review

Discover special offers, top stories, upcoming events, and more.

Thank you for submitting your email!

Explore more newsletters

It looks like something went wrong.

We’re having trouble saving your preferences. Try refreshing this page and updating them one more time. If you continue to get this message, reach out to us at with a list of newsletters you’d like to receive.