User guide

9-2 E
XTREME
W
ARE
S
OFTWARE
U
SER
G
UIDE
Q
UALITY OF
S
ERVICE
(Q
O
S)
O
VERVIEW OF
P
OLICY
-B
ASED
Q
UALITY OF
S
ERVICE
Policy-Based QoS allows you to protect bandwidth for important categories of
applications or specifically limit the bandwidth associated with less critical traffic. For
example, if voice–over-IP traffic requires a reserved amount of bandwidth to function
properly, using Policy-Based QoS, you can reserve sufficient bandwidth critical to this
type of application. Other applications deemed less critical can be limited so as to not
consume excessive bandwidth. The switch contains separate hardware queues on every
physical port. Each hardware queue is programmed by ExtremeWare with bandwidth
management and prioritization parameters. The bandwidth management and
prioritization parameters that modify the forwarding behavior of the switch affect how
the switch transmits traffic for a given hardware queue on a physical port.
The switch tracks and enforces the minimum and maximum percentage of bandwidth
utilization transmitted on every hardware queue for every port. When two or more
hardware queues on the same physical port are contending for transmission, the switch
prioritizes bandwidth use so long as their respective bandwidth management
parameters are satisfied. Switch products with the “i” chipset can be configured with up
to eight physical queues per port, while other Extreme switches can be configured with
up to four physical queues per port.
As with all Extreme Switch products, Policy-Based QoS has zero impact on
switch performance. Using even the most complex traffic groupings is “costless”
in terms of switch performance.
Policy-Based QoS can be configured to perform per-port Random Early Detection (RED)
and drop-probability. Using this capability, the switch detects when traffic is filling up
in any of the eight hardware queues, and performs a random discard on subsequent
packets, based on the configured RED drop-probability.
Instead of dropping sessions during times when the queue depth is exceeded, RED
causes the switch to lower session throughput. The destination node detects the
dropped packet, and, using standard TCP windowing mechanisms, slows the
transmission from the source node. RED drop-probability is configured on a
system-wide basis, and has a valid range from 0% to 100%. Only switches and modules
with the “i” chipset can use RED.