White Papers

average queue size determines when packets are marked for later transmission and when the transmission rate is reduced on an
interface during times of network congestion.
For example, in a best-effort network topology that uses WRED with instantaneous ECN, data packets may be transmitted at
a rate in which latency or throughput are not maintained at an effective, optimal level. Packets are dropped when the network
experiences a large traffic load according to the configured WRED thresholds. This best-effort network deployment is not
suitable for applications that are time-sensitive, such as video on demand (VoD) or voice over IP (VoIP) applications.
To resolve the problem of packet loss at times of network congestion, you may need to apply WRED with ECN and more finely
tune packet transmission for certain traffic types. To do so, you can configure the weight used to calculate the average queue
size; the average queue size is used to determine when to drop packets with WRED and when to mark packets with ECN when
WRED thresholds are exceeded.
The user-configurable weight in WRED and ECN provides better control in how the switch responds to congestion before a
queue overflows and packets are dropped or delayed. Using a configurable weight for WRED and ECN allows you to customize
network performance and throughput.
Setting Average Queue Size using a Weight
When a packet reaches a Z9500 switch enabled for WRED, the average queue size is computed. To determine the average
queue size, a weight factor is used. The weight factor is set to zero by default, which causes packets to be dropped by WRED
when a threshold is exceeded as follows:
● If the average queue size is less than the minimum WRED threshold, an incoming packet is queued.
● If the average queue size is greater than the maximum WRED threshold, an incoming packet is dropped.
● If the average queue size is between the minimum and maximum threshold values, the decision to drop or queue the packet
is based on the packet drop probability. The probability that a packet is dropped depends on the minimum and maximum
thresholds, and the mark-probability denominator. The rate of packet drop increases proportionally as the average queue size
increases, until the average queue size reaches the maximum threshold value. The mark-probability value is the number of
packets dropped when the average queue size reaches the maximum threshold value.
On the Z9500, you can configure the weight factor that determines the average queue size for WRED and ECN packet handling
by using the wred weight command.
The average queue size is computed using the last calculated average-queue size and the current queue size. The following is
the formula to calculate the average queue size: average-queue-size (t+1) = average-queue-size (t) + (current-queue-length -
average-queue-size (t))/2^N
where t is the time or the current instant at which average queue size is measured, t+1 is the next calculation of the average
queue size, and N is the weight factor.
In a topology in which network congestion varies over time, you can specify a weight to enable a smooth, seamless averaging of
packets to handle the bursty nature of packets based on the previous time sampling performed. You can specify a weight value
for front-end and backplane ports separately. The range of weight values is from 0 to 15.
You can enable WRED with ECN capabilities per queue to fine-tune packet transmission. You can disable WRED with ECN per
queue while configuring the minimum and maximum buffer thresholds for each WRED color-coded profile. You can configure
the maximum drop-rate percentage for yellow and green profiles. You can configure these parameters for both front-end and
backplane ports.
Global Service-Pools for WRED with ECN
You can enable WRED with ECN to work with global service-pools. Global service pools that function as shared buffers are
accessed by multiple queues when the minimum guaranteed buffers for a queue are consumed. The Z9500 switch supports four
global service-pools in the egress direction.
Two types of service-pools are used: one for lossy queues and the other for lossless (priority-based flow control (PFC)) queues.
NOTE: Service pool 1 for lossless queues is not supported in software releases that do not support PFC.
You can define WRED profiles and a weight on global service-pools for both lossy and lossless (PFC) service-pools. The
following events occur when you configure WRED with ECN on a global service-pool:
● If WRED/ECN is enabled on the global service-pool with threshold values and if it is not enabled on the queues, WRED/ECN
are not effective based on global service-pool WRED thresholds. The queue on which traffic is scheduled must have
WRED/ECN settings enabled for WRED to be valid for its traffic.
668
Quality of Service (QoS)