18.
Quality of Service
MN700004 Rev 01
215
respectively, from each queue. These are the byte counts you would specify in your custom
queuing configuration.
To determine the bandwidth distribution this ratio represents, first determine the total number
of bytes sent after traffic serviced on all of the three queues:
(1 x 1086) + (12 x 291) + (2 x 831) = 1086 + 3492 + 1662 = 6240
Then determine the percentage of the total number of bytes sent from each queue:
1086/6240, 3492/6240, 1662/6240 = 17.4, 56, and 26.6 percent
As you can see, this is close to the desired ratio of 20/60/20.
If the actual bandwidth is not close enough to the desired bandwidth, multiply the original
ratio of 1:11.2:1.3 by the best value, trying to get as close to three integer values as possible.
Note that the multiplier you use need not be an integer. For example, if we multiply the ratio
by two, we get 2:22.4:2.6.We would now send two 1086-byte packets, twenty-three 291-byte
packets, and three 831-byte packets, or 2172/6693/2493, for a total of 11,358 bytes. The
resulting ration is 19/59/22 percent, which is much closer to the desired ratio that we
achieved.
Technique for Avoiding QoS Congestion
The QoS congestion avoidance technique strives to avoid congestion by monitoring network
traffic loads at network and internetwork bottlenecks. In conditions of congestion, this
technique provides preferential treatment for premium class traffic in order to maximize
network throughput and capacity utilization while minimizing packet loss and delay.
Weighted Random Early Detection (WRED) Mechanism
One of the queuing schemes that have lately been gaining prominence is the WRED. Unlike
other queuing schemes, it is designed to prevent congestions beforehand, rather than to
manage them once they occur. WRED overcomes a situation known as
tail drop
, which
occurs when a burst of packets saturates a switch’s or router's buffer causing the last few
packets in the burst to be dropped because the buffer has no room left for them.
A
tail drop
is bad because it prevents you from controlling which packets are dropped.
Ideally, you would choose to save your high priority packets and allow your low priority
packets to be dropped. However, when the buffer is full, packets are lost before you can
identify their priorities.
WRED copes with this situation by watching the queues and starting to drop packets when the
queues begin to fill up. Dropping a packet or two doesn't save much space, but if a TCP
packet is dropped, TCP will throttle down its transmission rate, helping to keep things nice
and smooth.
WRED bases its decision about which packet to drop on IP Precedence. It does not check the
protocol (i.e. TCP versus UDP). Since most VoIP installations grant higher priority to their
VoIP traffic, which is typically UDP, WRED is more likely to drop the lower-priority TCP
traffic.
However, if you have a substantial amount of other UDP traffic, you should increase its
priority as well, so that your UDP traffic would not be dropped. If your UDP packets are
discarded, not only will you lose the packets (which must be retransmitted by the application,
if they are retransmitted at all), but also the ability of WRED to prevent the congestion will be
