Cisco CATALYST 4000 32 PORT 10/100 L3 MODULE WITH 2 PORT GBIC Guia De Especificação
limiting, due to the inherent windowing scheme used in flow control. You can adjust the
burst size parameter or rate parameter to obtain the required throughput.
burst size parameter or rate parameter to obtain the required throughput.
Q. What is the typical value of burst size to be used for rate-limiting on Layer 3 (L3)
switches?
switches?
A. L3 switches implement an approximation of the single token bucket algorithm in
firmware, and a reasonable burst size for the range of traffic rates is about 20,000 bytes.
The burst size should be chosen to include at least one maximum-size packet. With each
arriving packet, the policing algorithm determines the time between this packet and the
last packet, and calculates the number of tokens generated during the elapsed time. It
then adds this number of tokens to the bucket and determines whether the arriving
packet conforms to or exceeds the specified parameters.
firmware, and a reasonable burst size for the range of traffic rates is about 20,000 bytes.
The burst size should be chosen to include at least one maximum-size packet. With each
arriving packet, the policing algorithm determines the time between this packet and the
last packet, and calculates the number of tokens generated during the elapsed time. It
then adds this number of tokens to the bucket and determines whether the arriving
packet conforms to or exceeds the specified parameters.
Q. How does the input or ingress classification work?
A. Four hardware queues are supported on the egress of a port. Packets are classified by
input based on the three IP precedence bits, where the Least Significant Bit (LSB) is a
"don't care." See this table:
input based on the three IP precedence bits, where the Least Significant Bit (LSB) is a
"don't care." See this table:
Input classification is not supported for non-IP protocols. No input scheduling algorithm
is supported on the input besides FIFO.
is supported on the input besides FIFO.
Q. How does the output or egress scheduling work?
A. The egress side of the interface has four hardware queues, as described in
How does
the input or ingress classification work?
. When there is congestion, the packets are
transmitted on the outgoing interface based on the Weighted Round-Robin (WRR)
algorithm between the four hardware queues. Bandwidth is not explicitly reserved for
these four queues. Each of them is assigned a different WRR-scheduling weight, which
determines the way the queues share the interface bandwidth. The WRR weight is user-
configurable; you can assign a different WRR weight for each queue. The default values
are shown in the table in
algorithm between the four hardware queues. Bandwidth is not explicitly reserved for
these four queues. Each of them is assigned a different WRR-scheduling weight, which
determines the way the queues share the interface bandwidth. The WRR weight is user-
configurable; you can assign a different WRR weight for each queue. The default values
are shown in the table in
How does the input or ingress classification work?
. The higher
the WRR weight, the higher the effective bandwidth for that particular queue.
Q. Can the QoS output scheduling be changed on an interface level?
A. Yes, Weighted Round-Robin (WRR) scheduling can be configured at a system level
and on an interface level. The interface-level configuration overrides the system-level
configuration for that specific interface.
and on an interface level. The interface-level configuration overrides the system-level
configuration for that specific interface.
Q. Does the Weighted Round-Robin (WRR) work on an interface configured to be
in a bridge group?
in a bridge group?
IP
Precedence
Queue
Selected
Default Weighted Round-
Robin (WRR) Weight
000 & 001
0
1
010 & 011
1
2
100 & 101
2
3
110 & 111
3
4
Page 5 of 7
Cisco - Catalyst G-L3 Series Switches and WS-X4232-L3 Layer 3 Modules QoS FAQ
7/20/2004
http://www.cisco.com/warp/customer/473/125.html