FAST TCP

from Wikipedia, the free encyclopedia

FAST TCP (also FastTCP ) is an algorithm for congestion control for the Transmission Control Protocol (TCP) . FAST TCP is specially designed for high-speed networks with high transmission latency and was developed at the Netlab of the California Institute of Technology . It is currently being sold commercially by FastSoft. FastSoft was acquired by Akamai Technologies in 2012 .

FastTCP is compatible with existing TCP algorithms and only requires modifications to the data transmitting computer .

designation

The term FAST is a recursive acronym for F AST A QM S calable T CP , where AQM stands for A ctive Q ueue M anagement and TCP for T ransmission C ontrol P rotocol .

functionality

The Congestion Control (Engl. Congestion Control ) attempts the bandwidth of a network optimally use and at the same time to avoid overloading by regulating the transmission speed of the connected network components. In contrast to other algorithms such as TCP Reno , which use the loss probability of data packets in the network as a signal for overloading ( loss probability ), FAST TCP, like TCP Vegas , uses the delay in processing the packet buffer ( queuing delay ).

Most modern congestion control algorithms reduce the flow of data as soon as they discover that packets are being lost on the network. The average transmission rate thus depends on the packet loss rate. This approach has two disadvantages. First of all, low loss probabilities are required in order to obtain high transmission rates. In the case of TCP Reno, a very low loss probability is necessary, but newer algorithms such as H-TCP , BIC TCP and HSTCP also require lower loss rates than those of most wireless wide area networks . In addition, a packet loss only briefly generates a small amount of information about the network load, whereas the delay is a continuous value and generally reveals more information about the network.

FAST TCP tries the number of packets in queues (engl. Queue ) to hold over the entire network constant. The number of stored packets is estimated from the measured difference between the observed packet round- trip time (RTT) and the base RTT . The base RTT is defined as the round-trip packet time in the network without packet buffering and is estimated with the lowest observed round-trip packet time of the connection. If too few packets are in the queue, the sending rate is increased, and if too many packets are in the queue, the sending rate is reduced. In this regard, FAST TCP is a direct successor to TCP Vegas.

The difference between TCP Vegas and FAST TCP is the way in which the sending rate is adjusted if the number of stored packets is too high or too low. TCP Vegas uses a fixed step size, regardless of how far the current sending rate is from the target rate. FAST TCP, however, adjusts the correction increment to the distance of the network system from equilibrium ( equilibration , Eng. Equilibrium ) to. That is, the sending rate is adjusted in larger increments as the system is further from equilibrium, and in smaller increments as it is close to equilibration. This procedure improves the convergence speed and the stability.

Strengths and weaknesses

Delay-based algorithms can in principle keep a constant window size and thus avoid the oscillations of packet loss rate-based algorithms. They also detect network congestion sooner, since the delay time already in partially filled buffers changes, but packet losses occur only when completely filled buffers. This can be both a strength and a weakness. If only delay-based algorithms are used in a network, the inefficiency of loss-rate-based algorithms can be avoided. However, if both types of algorithms are used in a network, delay-based algorithms tend to be comparatively less aggressive. However, this can be avoided by using appropriate parameters, which leads to the complex interactions described by Tang et al. have been described.

The measurement of the delay times can be obtained by jitter are distorted by the operating system - Scheduling or the bus caused compounds.

An ns-2 simulation has shown that existing FAST TCP flows can falsify the approximation of the base RTT via the base network delay ( round-trip propagation delay or RTPD) of subsequent FAST TCP flows. This misjudgment of the network runtime has the effect that newer FAST TCP flows act more aggressively and achieve a higher data throughput, which, however, disadvantages existing flows. The authors of the simulation suggest a solution to this problem.

Whether the strengths or the weaknesses of FAST TCP predominate is not clear and depends very much on the application scenario.

Generalized FAST TCP

FAST TCP has shown great promise in terms of system stability, data throughput and fairness. However, a data buffer is required whose size increases linearly with the number of overloaded flows in a connection. The Yuan et al. Generalized FAST TCP algorithm proposed as an extension of FAST TCP achieves (α, n) -proportional fairness in the stable connection state. In addition, the requirements for the buffer size are reduced to the nth power of the flow number.

Intellectual property

Unlike most other TCP congestion control algorithms, FAST TCP is protected by several patents. Instead of striving for standardization via the IETF , the inventors of FAST TCP, Steven H. Low and Cheng Jin, are trying to commercialize the algorithm via the company FastSoft.

See also

Web links

Individual evidence

  1. ↑ Jeff Young: Akamai Acquires FastSoft . September 13, 2012. Retrieved September 13, 2012.
  2. ^ Barons Nick, Jin, Cheng; Low, Steven H. and Hegde, Sanjay: FAST TCP: motivation, architecture, algorithms, performance Archived from the original on September 6, 2006. In: IEEE / ACM Trans. On Networking . 14, No. 6, 2006, pp. 1246-1259. doi : 10.1109 / TNET.2006.886335 .
  3. ↑ Cheng Jin, D. Wei, SH Low, J. Bunn, HD Choe, JC Doyle, H. Newman, S. Ravot, S. Singh, F. Paganini, G. Buhrmaster, L. Cottrell, O. Martin, Wu -Chun Feng: FAST TCP: from theory to experiments Archived from the original on May 12, 2006. In: IEEE Network . 19, No. 1, 2005, pp. 4-11. doi : 10.1109 / MNET.2005.1383434 .
  4. ↑ a b Ao Tang, Wang, Jiantao; Low, Steven H. and Chiang, Mung: Network Equilibrium of heterogeneous congestion control protocols . In: IEEE INFOCOM . March 2005.
  5. ↑ a b L. Tan, C. Yuan, and M. Zukerman, “FAST TCP: fairness and queuing issues,” IEEE Commun. Lett., Vol. 9, no. 8, pp. 762-764, Aug 2005.
  6. ↑ Cao Yuan, Liansheng Tan, Lachlan LH Andrew, Wei Zhang, Moshe Zukerman: A Generalized FAST TCP scheme . In: Computer Communications . 31, No. 14, 2008, pp. 3242-3249.
  7. ↑ Patent US9253104 : Dynamic Adjustment Of Receive Window Utilized By A Transmitting Device. Filed August 14, 2014 , published April 18, 2014 , applicant: AKAMAI TECH INC, inventor: JIN CHENG [US]; LEE GEORGE S [US]; LOW STEVEN [US]; NG DARREN [US]; WITT RYAN [US]. ‌
  8. ↑ Patent US7974195 : Method and apparatus for network congestion control. Filed May 14, 2004 , published June 5, 2011 , applicant: CALIFORNIA INST OF TECHN [US], inventor: JIN CHENG [US]; LOW STEVEN [US]; WEI XIAOLIANG [US]. ‌