10-9 Weighted Random Early Detection (WRED) | Cisco Field Manual[c] Router Configuration

WRED monitors traffic loads and performs selective packet dropping as a way to avoid and prevent congestion.
WRED uses weight values to differentiate between classes of service, providing better service to higher-priority traffic.
WRED is aware of IP Precedence, delivering more higher-priority packets and discarding more lower-priority packets. RSVP flows are also preferred while dropping packets from other traffic flows.
WRED works well with TCP/IP, causing TCP to retransmit a dropped packet at a reduced rate. By randomly dropping packets, TCP flows can be reduced in a nonglobal fashion.
Flow-based WRED keeps track of traffic flows through an interface. Flows that try to monopolize the network are penalized by having more packets dropped.
If WRED is not configured, tail drop congestion avoidance is used. When queues fill during congestion, packets from all flows are dropped until the congestion lessens. Traffic is treated equally, with no random or preferential packet drops being performed.

Configuration

Use WRED on an interface.
1. Enable WRED on the interface:
```
 (interface)  random-detect  [  prec-based   dscp-based  ] 
```
  WRED can be based on IP Precedence ( prec-based, the default) or on IP DSCP ( dscp-based ).
2. (Optional) Set the WRED thresholds:
```
 (interface)  random-detect  {  precedence   precedence   dscp   dscp  }  min-threshold  max-threshold mark-prob-denominator 
```
  WRED can be based on IP Precedence ( precedence ) or IP DSCP ( dscp ). When the average queue length reaches the minimum threshold, min-threshold (1 to 4096 packets), some packets with the precedence value or the dscp value are dropped. Likewise, when the queue length reaches max-threshold ( min-threshold to 4096 packets), all packets with the precedence value are dropped.
  
  When the queue meets the threshold level, one out of every mark-prob-denominator packets (1 to 65536; the default is 10 packets) is dropped.
  
  IP Precedence values are given as precedence (0 to 7), and DSCP values are given as dscp (0 to 63 or a keyword: ef, af11, af12, af13, af21, af22, af23, af31, af32, af33, af41, af42, af43, cs1, cs2, cs3, cs4, cs5, or cs7.
Use flow-based WRED instead.
1. Enable flow-based WRED on an interface:
```
 (interface)  random-detect flow  
```
2. Adjust the average depth factor for flow queues:
```
 (interface)  random-detect flow average-depth-factor   scaling-factor  
```
  If a flow queue becomes nonadaptive to congestion, you can change the scaling factor that WRED uses to compute the queue depth. The scaling-factor ( 2, 4, 8, or 16; the default is 4) multiplied by the average flow depth should equal the peak flow depth.
3. Adjust the flow count for the interface:
```
 (interface)  random-detect flow count   count  
```
  WRED keeps a count of the active flows through an interface. Based on this count, buffers are allocated. The count (16 to 32768; the default is 256) can be adjusted to tune the buffer utilization if you anticipate more flows.

Weighted Random Early Detection Example

WRED is configured on serial interface 3/1 based on IP Precedence. The default WRED thresholds are used. Serial interface 3/2 has flow-based WRED configured because there are no known criteria for IP Precedence or DSCP being used. A large number of active flows is not expected, so the WRED flow count has been reduced to 128.

  interface serial 3/1   ip address 192.168.204.33 255.255.255.0   random-detect prec-based   interface serial 3/2   ip address 192.168.205.33 255.255.255.0   random-detect flow   random-detect flow count 128