Random early detection

Random early detection (RED ), also known as random early discard or random early drop is an algorithm for a network scheduler for congestion avoidance (see overload avoidance algorithm).

In conventional tail drop algorithm, a router or other network component buffers as many packets as possible and discards those that are no longer fit into the queue. If the queue is full, it's called overloaded the network. The tail drop algorithm distributes the queue slot does not do justice between the currents of the network nodes. Tail Drop can also lead to TCP Global Synchronization with all TCP connections at the same time and wait begin simultaneously after their timeouts again to send. The networks are therefore not optimal either busy or overloaded.

Here RED comes into play. It monitors the average queue length and discards (or marked if it is used in conjunction with ECN) packets based on statistical probabilities. If the queue is almost empty, all incoming packets are accepted. When growth of the queue, the probability of rejecting incoming packets grows. If she - is ( almost see maximum upper limit ) fully, the probability of discarding 1 and, accordingly, all packets are dropped.

RED provides greater fairness than the tail drop algorithm. The more communicates a host, the more likely it is that its packets are dropped is. Early Detection helps prevent TCP Global Synchronization.

RED takes no account of traffic classes with Quality of Service ( QoS). Weighted RED ( WRED ) and Red In / Out ( RIO) provide Early Detection in material respects to QoS.

RED is a way to avoid network congestion. RED- enabled network components use a port queue. Are more of these buffers per port available, WRED can be ( Weighted Random Early Detection, weighted random early detection ) are used. RED informs the transmitter and receiver only indirectly through the jam by it (usually 1-2 %) deletes some randomly selected packets from the queue when it reaches a certain level (eg 50 %). With increasing length of the buffer exponentially more packets are dropped, the more the average queue length of the 100% mark approaches. The average level is usually computed over the duration of one second. Due to the randomized approach of RED it comes automatically to a fair sharing of bandwidth.

Other variants

Weighted RED ( WRED ) can assign different probabilities for different priorities and queues.

Adaptive / Active RED ( ARED ) considers the average queue length and then decide how aggressive RED is to take action. If the average queue length is varied to the lower limit, then the Early Detection method is too aggressive. When the average queue length varies around the upper limit, Early Detection is too conservative. The algorithm changes the probability by referring to his ability to "feel" the aggressiveness of discarding traffic to.

672086
de