13 thoughts on “FastSoft Tweaks TCP to Speed the Internet”

  1. There is a far simpler, less math-vomit-inducing way to explain what they’ve done.

    If you reduce the distance between the sender and the recipient you can increase the transmission rate. By introducing a box on the sender site that essentially gateways the SYN/ACKs other aspects of the TCP stream they have (or they should have) just convinced the sender to fire things off as fast as the wire can allow. They, in turn, manage the connection with the real client who is quite a longer distance (in terms of latency) away.

    This is not new technology as far as I’m aware. I think I started doing thins like this back when running SYN proxies became the norm on massive application infrastructures.

    Then again, I’m sure it’s a good speed tweak and makes a difference. I wouldn’t call it a new kind of TCP though. For that to happen, both sides have to be willing to talk differently. 🙂

  2. It’s fair to say that the vast majority of traffic on the Internet uses TCP but Skype for sure does not (at least not for carrying voice packets). VoIP services are typically built on top of UDP.

  3. Hi David

    Thanks for pointing out several subtleties of the technology.

    1. You are absolutely right that the idea of proxying a TCP connection is not new, and it is indeed an effective mechanism through which to address many issues that arise in massive infrastructures, as you pointed out. By inserting an appliance between the sender and the Internet, it gives us a way to deal with fluctuations in the Internet and maintain end-to-end application performance. The innovation here is not in convincing the sender to send fast, but the various algorithms to manage network fluctuations effectively to deliver robust performance to the clients. This include algorithms to optimally utilize available bandwidth in Internet, to be extremely resilient to packet loss, to maintain throughput, robustness, and fairness across long distances and heterogeneous receivers.

    2. You are also right in pointing out that the basic TCP framework is intact in our approach. We have, however, completely re-designed the *implementation* of key functions in TCP that affect content delivery over the Internet, so that it is more efficient, robust, and evolvable in today’s networks and for today’s applications. We have implemented all these innovations in a way that is TCP-compliant – that’s why the appliance accelerates without having any hardware or software installation at the client side.


  4. Hi Steve,

    In your post, you mention algorithms you’ve developed that optimally utilize available bandwidth and reduce packet loss etc. I know that Internap’s MIRO technology and smart routing does this when you purchase Internap bandwidth. Is your technology complimentary to Internap’s devices like the FCP box or would it replace it?

  5. Does this work in both directions, or just for outgoing data?

    I am assuming just outgoing data, but I would like to see something in the reverse that helps speed up incoming data (We run a bunch of spiders with massive amounts of incoming data).

  6. Hi Guy

    I’m not familiar with the technologies that you mentioned, but in general, routing is layer 3 and transport (e.g. TCP) is layer 4 (at least conceptually; I know there are exceptions in implementation etc), so they complement each other, in that one attempts to choose the best path from point A to point B and one attempts to make the most effective use of the chosen path.

    The most remarkable ingenuity of the Internet pioneers, and one that is singularly responsible for the explosive growth of the Internet, is to have gotten the layering architecture right, of which the above is an example. Each layer provides a few functions well (media access, routing, congestion control & loss recovery, applications, etc) and each can evolve independently, unlocking innovations across almost all IT landscape and beyond. The very fact that multiple companies and industries are able to cooperate/co-exist/compete to build/operate/serve from/monetize on the same network infrastructure is an astounding testament of the architectural success of these pioneers, unimaginable before the Internet era. Can’t help but digress in awe…


  7. It is a pleasure to learn from the guys that invent ground breaking IP, and those that wrtie about it really well.
    Thanks Dr Low and OM

  8. There is a rumor that FastSoft beat Akamai for the Digital Asset Management
    Network business at Getty Images. Pretty impressive If true.

  9. Bruce — I obviously know nothing of the deal, but I’ll point out that it’s not hard to beat Akamai on price, and sometimes that’s the battlefield where a startup is able to get a foothold initially.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.