When Speed Counts – Making FTP Faster
Caltech is the proud academic home of the Jet Propulsion Lab and Palomar Observatory. A private and premier research institution, [[California Institute of Technology]] is also proud of having faculty earn 32 Nobel prizes. Caltech is also the originator of Fastsoft, the CalTech and VC-backed TCP accelerator.
I had a telephone briefing with Dan Henderson, the VP Marketing for Fastsoft, who gave me a low-down on the company.
There is no doubt that the venerable [[Transmission Control Protocol]] is an incredibly important protocol for all things Internet. The wiki points out that
"because TCP is optimized for accurate delivery rather than timely delivery, TCP sometimes incurs relatively long delays (in the order of seconds) while waiting for out-of-order messages or retransmissions of lost packets"
but, probably most important is that the designers of TCP worried about congestion and presented several mechanisms for controlling the network response in the face of evidence of network congestion. [[Congestion collapse]] is a condition when the network is fully congested, packets are routinely discarded and little useful traffic transpires. This actually happened in the Internet in 1986-8.
In the early days of the Internet, when having a 1 Mbps WAN link was rare, it wouldn't take a lot of data transmission to overwhelm the link. But, in a day when consumers regularly have 2 Mbps links and businesses can have Gbps service for what it cost 1 Mbps a decade ago, the risk of congestion and therefore the appropriate responses to that risk are different. That's why TCP acceleration and [[TCP congestion avoidance algorithm]] have been particularly important mechanisms under research.
The researchers at CalTech's NetLab were the originators of the [[Fast TCP]] which uses delay information instead of lost packets (which really rarely happen in the 100-Gbps fiber optic backbones we have today) to respond to the presence of network congestion.
And the results are impressive. Fastsoft consistently wins competitions and can deliver between 2 and 30 times faster file delivery.
It's an asymmetrical appliance with licensing based on expected throughput to be supported. Starting at approximately $10,000 for 10 Mbps, and going as high as $80,000 for 1 Gbps throughput (dealers may sell for less). Deploy one device in the data center and deliver significantly faster files to all TCP-bearing users, immediately. No need for remote devices. No need for system management software. An asynchronous design is a big deployment and economic advantage over other WAN accelerators such as offered by Riverbed or Expand Networks. It might be that the bit-caching techniques of the Steelhead appliance is a stronger fit for the specific application types used within that enterprise, but the overhead cost of rolling out multiple appliances into remote offices such as shipping, installing, configuring, powering, cooling and managing those appliances and the like can easily overcome any incremental performance benefit of the appliance model.
With a small direct sales force focused on media and entertainment markets that have a particular interest in getting large files to the people who ask for them, and a developing VAR network for international sales, Fastsoft is challenging the issues where [[Content Delivery Network]]s have been the only practical answer. Of course, symmetrical acceleration techniques won't work in these public service settings, but Fastsoft will.
Being optimized for large file transfers over long distances like video clips and Internet TV is important. Particularly given that video promises to be a big part of the demand for speed in the next 4 years.
This post has already been read 0 times!