High Speed TCP
The current proposal for TCP relies on the simple AIMD parameters of (1,0.5). In long latency, high bandwidth environments, these values are simply insufficient to make use of the available bandwidths available.
New implementations of TCP rely on the theory that the loss rate is proportional to the TCP throughput. By altering this 'response function' and fitting suitable parameters for (a,b), one is no longer limited by the restrictions of standard AIMD upon transfers and therefore can achieve high throughput transport with only sender side modifications.
Other implemenations rely on giving TCP more information about the network in order to correctly predict the required increase/decrease. This 'multibit' knowledge of the network is an adaption of the Vega protocol.
There are currently 3 main proposals:
| Description | |
| HSTCP | An adaptive algorithm that increases its additive increase parameter and decreases its decrease parameter in relation to the current congestion window size. |
| Scalable TCP | An algorithm based on HSTCP that utilises the fact that the response function can be altered to give constant recovery times in the even of loss/congestion. Decrease is an eighth. |
| FAST | Based on the Vegas core, but utilising the rtt information as a primary rather than a secondary congestion feedback signal; it uses delay information from its data packets and acks as an indication of congestion on the network. |
| Mon, 3 November, 2003 14:26 |   |
Room D14, High Energy Particle Physics, Dept. of Physics & Astronomy, UCL, Gower St, London, WC1E 6BT
