How to Transfer Protocol TCP congestion control?

Congestion Control

Congestion means that too many packets arrive at a certain part of the communication subnet, so that part of the network time to process, so as to cause this part of the whole network performance degradation phenomenon, serious and even cause network traffic to a halt, appears dead lock phenomenon. This phenomenon is often seen with the road network as traffic congestion, increased holidays, when a large number of vehicles in the road network, all kinds of traffic to interfere with each other, so that each vehicle time to reach a destination are relative increase (ie increased latency) and even sometimes on a certain road vehicles can not start due to clogging (ie localized deadlock).

Causes of congestion

1, mainly due to traffic, a plurality of outputs corresponding to the plurality of inputs, the inflow line has a plurality of packet arrival, and require the same output line, this time, if the router does not have enough memory to store all the packets, then some the packet will be lost.

2, with a slow processor router's sake, that it is difficult to complete the necessary processing, such as buffer queue, updating routing table.

Preventing congestion

How to Transfer Protocol TCP congestion control?

1, the transport layer may be employed: retransmission policy, scrambled caching policy, acknowledgment policy, and determines the flow control policy timeout policy.

2, the network layer may be: internal subnet virtual circuits and the data reported strategy, the packet queuing and service policy, packet discard policy management survival and packet routing algorithm.

3, the data link layer may be employed: retransmission policy, scrambled caching policy, acknowledgment policy and flow control policy.

Control Method

1, buffer allocation method: This method is used in a virtual circuit packet-switched network, the nodes in the network to establish a virtual circuit as a virtual one or more pre-allocated data buffer. If a node buffer has been filled, the call request packet to choose another route, or return a "busy" signal to the caller.

2, the packet drop method: This method does not have to retain the pre-buffer, when the buffer fills, the incoming packet is discarded. If subnetworks datagram service is provided, the method used to prevent the packet discarding congestion without causing a big impact. However, if the communication subnets virtual circuit service, the packet must be discarded to save backup somewhere, so as to solve the congestion to re-transmission. There are two solutions to be discarded packet retransmission method for a packet is to be discarded transmitting node times out, and re-transmits the packet until the packet is received; the other is to transmit the packet is discarded at a node after a certain number of attempts send to give up, and force data source node timeout and resume sending.

3, fixed control method: This method sets the appropriate number of special message called "permit" in the communication subnet, in some part of the license policy previously assigned to the respective subnetworks source node before the start of work, the other after part of the subnet to work around four in the net. When the source node wants to send packets from the source end of the system, it must first have a license, and each send a logout packet a license. After each side is the destination node receives a packet and submit it to the destination system, will generate a license.

Guess you like

Origin blog.51cto.com/13609234/2425184