x
Optical/IP

FastSoft Wants CDNs

A year after debuting to the media, FastSoft Inc. has amassed some noteworthy clients for its transmission control protocol (TCP) acceleration technology, but the company says it's focusing less on telcos and more on content delivery network (CDN) providers for now.

In part, that's because the company's claim to fame -- fixing TCP's ills -- has its most visible benefits with video delivery. The company is saying it can help CDNs improve video quality and thus lower customer churn.

FastSoft is going to be butting heads with Digital Fountain , which is also camped out at CDNs' doorsteps with a video-helping pitch. And it's still got some obvious competition from the WAN acceleration products out of Cisco Systems Inc. (Nasdaq: CSCO) and Riverbed Technology Inc. (Nasdaq: RVBD).

Today's announcement has FastSoft upping its speed to the gigabit level with the E 1000 appliance. (See FastSoft Goes Gigabit.) A 10-Gbit/s box is forthcoming, but FastSoft isn't putting a timeframe on its arrival.

FastSoft began life as a project funded by Cisco and the National Science Foundation (NSF) . The company made its first PR splash a year ago when it raised its initial venture funding, a $4 million sum. (See FastSoft Receives $4M and FastSoft.)

The company launched in early 2006, after rebuffing an acquisition offer by Cisco, according to FastSoft vice president of marketing Dan Henderson. "Cisco came along and offered something, but there was more opportunity in creating a company than selling it to Cisco," he says.

FastSoft's calling card is FastTCP, an algorithm that boosts TCP performance by reacting to network slowdowns more immediately and more subtly.

The problem is that TCP can't tell the network is congested until it starts noticing lost packets -- a late symptom -- and then reacts by cutting the transmission speed drastically. FastSoft picks up on congestion by watching for packet delays, and it reacts more gently.

TCP has behaved like this for 21 years, but it "was not much of a problem in the past because the link speeds were very slow," says Burton Group analyst Dave Passmore. "Now, it becomes a much bigger deal because all it takes is a bit error to kill your throughput. That's what the WAN optimizers are trying to address, except they need boxes at both ends."

And that's the advantage FastSoft claims it's got over Riverbed and Cisco. Unlike WAN acceleration, FastSoft only needs equipment installed at one end, making FastTCP an option for widely distributed services like video broadcasts.

So who's using this stuff? It turns out FastSoft has built up some star power on its client list, including Walt Disney Co. (NYSE: DIS)/ESPN, Siemens AG (NYSE: SI; Frankfurt: SIE), and, as Light Reading has learned, Getty Images (which, coincidentally, gets hired by Light Reading for the occasional fancy photo shoot).

Telcos and service providers could make use of FastTCP too, but FastSoft is concentrating on CDNs for now. The company doesn't have any wins in that space yet but is "very close to announcing several," Henderson claims.

— Craig Matsumoto, West Coast Editor, Light Reading

mr zippy 12/5/2012 | 3:32:52 PM
re: FastSoft Wants CDNs "The problem is that TCP can't tell the network is congested until it starts noticing lost packets -- a late symptom -- and then reacts by cutting the transmission speed drastically. FastSoft picks up on congestion by watching for packet delays, and it reacts more gently.

TCP has behaved like this for 21 years, but it "was not much of a problem in the past because the link speeds were very slow," says Burton Group analyst Dave Passmore. "Now, it becomes a much bigger deal because all it takes is a bit error to kill your throughput. That's what the WAN optimizers are trying to address, except they need boxes at both ends."

Seems this analyst isn't up with what's happened with TCP in the last 21 years - hello TCP Explicit Congestion Nofication

http://en.wikipedia.org/wiki/E...

Also, there's been plenty of work on congestion control algorithms for TCP - have a look at "Other TCP congestion avoidance algorithms" under:

http://en.wikipedia.org/wiki/T...



Pete Baldwin 12/5/2012 | 3:32:51 PM
re: FastSoft Wants CDNs Yes, there's been a ton of work in this area. Passmore is aware of it -- he was also telling me about Microsoft tweaking TCP in Vista, something that could (tangent alert) cause trouble for WAN optimization gear, since a faster TCP in Windows would mean less perceived need for speeding up the network...

Anyway. Passmore was talking more about the general case: Why didn't people care so much about TCP's shortcomings back in 1980something.

You bring up an excellent point, though. You can consider FastSoft as just the latest (and not the cheapest) enhancement available to TCP, one of many many alternatives out there. Is that, by itself, going to be enough to build a company? I applaud them for giving it a shot, but we all know that good technology isn't always enough.
rpm23 12/5/2012 | 3:32:45 PM
re: FastSoft Wants CDNs any improvements in tcp are good for video file downloads.
I am not sure it makes much sense, in the first place, to transport streaming video over TCP with its congestion control algorithm designed for elastic data services. Fixing the congestion control algorithm helps but the basic arrangement is flawed IMHO. what you should do is transport over UDP and do one of two things (or both) when enough network bandwidth is not available. Drop less important packets (eg., p/b frame related packets in mpeg) from the stream in the network and/or adapt the source *encoding* to the network bandwidth (if you can somehow get a feedback of that or estimate it). keeping the encoding constant and adjusting the TCP transmission rate (as any TCP scheme does) is just going to result in random packet drops at the TCP buffers.
HOME
Sign In
SEARCH
CLOSE
MORE
CLOSE