3Com Router User Manual

							50
CONGESTION AVOIDANCE
This chapter covers the following topics:
■Congestion Avoidance Overview
■WRED Configuration
■Displaying and Debugging Congestion Avoidance
■Congestion Avoidance Configuration Example 
Congestion Avoidance 
OverviewThe purpose of the congestion avoidance technology is to monitor the network 
traffic flow, predict the congestion and effectively prevent the congestion 
occurring at the bottleneck of the network. In a number of the congestion 
avoidance mechanisms, Random Early Detection (RED) technology is widely used.
Excessive congestion can create damage on the network resource, and measures 
must be taken to avoid it. Here, the so-called congestion avoidance refers to a 
traffic control mechanism that, by monitoring the usage of the network resources 
(such as the queue or memory buffer), removes the network overload by dropping 
packets on its own initiative to adjust the network traffic in case of the network 
congestion.
Compared to the end-to-end flow control, steam control here has wide-range 
meaning, it affects more service steam load in the router. Of course, when the 
router discards the packet, it does not reject the cooperation with the flow control 
action, such as the TCP flow control, of the source end, so as to adjust the traffic 
of the network to a rational load status in a more efficient way. The combination 
of a good drop policy and source end flow control mechanism always pursue the 
maximization of the network throughput and service efficiency and the 
minimization of the packet drop and delay.
Traditional Drop Policy The traditional drop policy utilizes the tail-drop method. The tail-drop applies to all 
the traffic flow. It can not distinguish the service level. During the occurrence of 
the congestion, the data packet of the queue tail will be dropped, until the 
congestion is settled.
The host running the TCP protocol responds to numerous drops by reducing the 
packet transmission rate. When the congestion is cleared, the transmission rate of 
the data packet is increased. In this way, tail-drop can cause the TCP Global 
Synchronization. When the queue drops multiple TCP packets simultaneously, it 
causes multiple TCP connections to come into congestion avoidance and slow 
startup states simultaneously, and reduces and adjusts the traffic at the same time, 
then the traffic peak occurs as the same time as the reduction of the congestion,  
						

							698CHAPTER 50: CONGESTION AVOIDANCE
and it causes the sudden increase and decrease of the network traffic, and the line 
traffic always fluctuates between the states of few or none and full.
RED and WREDRED and WRED can avoid global synchronization of TCP by dropping packets 
randomly. When the packets of a TCP connection are dropped, and transmission 
slows down, other TCP connections can still send packets at high rates, thus 
improving the utilization of the bandwidth.
RED and WRED avoids the TCP global synchronization phenomenon through the 
random drop packets--when the packet of a TCP connection is dropped and the 
transmission speed is reduced, other TCP connections still have the higher 
transmission speeds. Thus, it is always the case that some TCP connection 
performs the faster transmission, increasing the use ratio of the line bandwidth.
Both RED and WRED compare between the queue length, and minimum and 
maximum thresholds, to perform the drop (this is to set the absolute length of the 
queue). It will cause the unfair treatment on the burst data flow and be 
disadvantageous for the transmission of the data flow. Therefore, when 
comparing the minimum and maximum thresholds, and when dropping, the 
average lengths of the queue are adopted (this is to set the relative value of the 
comparison between the queue threshold and the average length). The average 
length of the queue is the result of the low pass filtering of the queue length, it 
reflects the variation trend of the queue, and is not sensitive to the burst change 
of the queue length, so as to avoid the unfair treatment on the burst data flows.
The relationship between WRED and queue mechanism is shown in Figure 223
Figure 223   Schematic diagram of the relationship between WRED and queue mechanism
In the RED class algorithm, a pair of minimum threshold and maximum threshold is 
set for each queue, and the following specification is set:
■When the length of the queue is less than the minimum threshold, no packet is 
dropped.
■When the length of the queue is larger than the maximum threshold, all 
incoming packets are dropped.
incoming packets
queue1 weight1
queue2 weight2
classify
outgoing packets
queueN-1 weightN-1
queueN weightN
transmit
queueinterface
¡ -¡-
scheduler
¡
-¡- WRED drop
Discarded
packets 
						

							WRED Configuration699
■When the length of the queue is between the minimum threshold and 
maximum threshold, the WRED algorithm is used to calculate and determine 
whether the packet is dropped. The specific method is that each incoming 
packet is allocated with a random number, which is compared with the drop 
probability of the current queue, if it is larger than the drop probability, the 
packet is dropped. The longer the queue is, the higher the drop probability 
is--but there is a maximum drop probability.
When WRED and WFQ is cooperated, the flow based WRED can be implemented. 
During the classification, different flows have their own queues, for the flow with 
small traffic, as its queue length is always smaller, the drop probability will be 
smaller, too. However, as the flow with large traffic will have larger queue length, 
more packets are discarded, protecting the benefit of the flow with smaller traffic.
Different from RED, the random number generated by WRED is based on the IP 
priority, it considers the benefit of the high priority packets, and relatively reduce 
the drop probability of the high priority packets. The 3Com router takes WRED as 
its congestion avoidance policy.
WRED ConfigurationWRED configuration includes:
■Enable the WRED Function of the Interface  
■Configure Weight Factors when Calculating WRED Average Queue Length 
■Set the Priority Parameters for WRED
Enable the WRED 
Function of the Interface WRED must first be enabled, and then other parameters related to WRED can be 
configured.
Please perform the following configurations in the interface view.
Ta b l e 738   Enable WRED
By default, the system disables WRED so the queue avoids congestion by using the 
tail-drop policy.
WRED can only operate with WFQ, and cannot be used separately or coperated 
with other queue. Therefore, before the startup of WRED, WFQ must have been 
applied to the interface.
Enabling WRED can be effective only in all physical interfaces, while this command 
is ineffective in the logic interface.
Configure Weight 
Factors when 
Calculating WRED 
Average Queue LengthPlease perform the following configurations in the interface view.
Ta b l e 739   Configure the WRED Weighted Factor for Calculating the WRED Average 
Queue Length
OperationCommand
Enable the WRED function of the interfaceqos wred
Disable the WRED function on the 
interfaceundo qos wred
OperationCommand 
						

							700CHAPTER 50: CONGESTION AVOIDANCE
exponent is the filtering coefficient for calculating the average queue length, and 
the range of the value is 1 to 16, and the default value is 9.
When exponent=0 and the queue length exceeds the threshold, WRED will act 
accordingly. When exponent is higher, WRED will act slowly to the change of 
queue status.
This configuration should be performed after enabling WRED in the interface view.
Set the Priority 
Parameters for WREDYou can set WRED drop lower threshold value, upper threshold value, and drip 
probability denominator according to packet priority. The reciprocal value of the 
denominator discard-prob will be taken as the maximum drop probability. The 
system will handle the queues according to the length of the queues.
■If the queue length is lower than the low-limit, no packet will be dropped.
■If the queue length is between low-limit and high-limit, the drop probability 
will increase with the queue length till it is almost equal to the reciprocal value 
of discard-prob. 
■If the queue length is equal to or greater than high-limit, all the packets will be 
dropped.
Please perform the following configurations in the interface view.
Ta b l e 740   Configure the Related Parameters for the Packets of Specific IP Priority 
ip-precedence is the IP precedence, and the range of the value is 0 to 7.
low-limit and high-limit are the minimum and maximum thresholds respectively. 
The default values are 10 and 30 respectively, and the range of the value is 1 to 
1024.
discard-prob is the drop probability denominator and its default value is 10. The 
reciprocal of discard-prob will be the maximum drop probability. The range of this 
parameter is 1 to 255.
It should be noted that this configuration can only be performed after WRED is 
enabled in interface view.
Configure the WRED weighted factor for 
calculating the WRED average queue 
length.qos wred weighting-constant exponent
Recover the default value of the WRED 
weighted factor for calculating the WRED 
average queue length.undo qos wred weighting-constant
OperationCommand
Configure the related parameters for the 
packets of specific IP priority qos wred ip-precedence ip-precedence 
low-limit low-limit high-limit 
high-limit discard-probability 
discard-prob
Recover the default values of the related 
parameters for the packets of specific IP 
priorityundo qos wred ip-precedence 
ip-precedence 
						

							Displaying and Debugging Congestion Avoidance701
Displaying and 
Debugging 
Congestion AvoidanceTa b l e 741   Display and Debug Congestion Avoidance
Congestion Avoidance 
Configuration 
Example 
1Configure a WFQ queue.
[Router] interface ethernet 0
[Router-Ethernet0] qos wfq
2Enable WRED.
[Router-Ethernet0] qos wred
3Configure the exponent to calculate the average WRED queue length.
[Router-Ethernet0] qos wred weighting-constant 1
4Configure the lower threshold, upper threshold, and drop probability denominator 
of the WRED queue with precedence 0 to be 10, 1024 and 30 respectively.
[Router-Ethernet0] qos wred ip-precedence 0 low-limit 10 high-limit 
1024 discard-probability 30
OperationCommand
Display the WRED configuration 
conditions and statistic information of the 
interfacedisplay qos wred [ interface type 
number ] 
						

							702CHAPTER 50: CONGESTION AVOIDANCE 
						

							XII
DIAL-UP
Chapter 51Configuring DCC
Chapter 52Configuring Modem 
						

							704 
						

							51
CONFIGURING DCC 
This chapter covers the following topics:
■DCC Overview
■Configuring DCC 
■Displaying and Debugging DCC
■DCC Configuration Examples
■Troubleshooting DCC
DCC OverviewDial Control Center (DCC) is the routing technique adopted when the routers 
interconnect via a PSTN (Public Switched Telephone Network) or ISDN (Integrated 
Services Digital Network). In DCC, the routers are interconnected through PSTN.  
The connections are established through dialing when data transmissions are 
required. A DCC dialing is required to set up a link for transmitting information. 
When the link becomes idle, the link established by DCC will be automatically 
disconnected.
Under certain circumstances, routers establish connections for communications to 
satisfy specific requirements. Therefore, the data transmission are 
time-independent, burst and in small size. DCC provides a flexible, economical, 
and efficient solution for such applications. In practice, DCC guarantees the 
priority of communications through designated backup lines. In the case that a 
primary line for normal communications become unavailable for any reasons, DCC 
uses the designated backup channels to carry out the communications to assure 
the required services are timely completed.
Frame Relay network through a leased line. To reduce the cost, you can adopt 
frame relay over ISDN to access the frame relay network through ISDN line. 
Meanwhile, ISDN network can act as the backup of frame relay network.
Terms in DCC 
ConfigurationThe following terms are commonly used in DCC configurations:
■Physical interface: The physical interface that actually exists, like the serial, BRI, 
asynchronous, and AM interfaces.
■Dialer interface: Logical interface set for configuring DCC parameters. A 
physical interface can inherit the DCC configuration after it is bound to the 
dialer interface.
■Dial interface: A general term describing an interface for dialup connection. It 
can be a dialer interface, a physical interface bound to the dialer interface, or a 
physical interface directly configured with DCC parameters. 
						

							706CHAPTER 51: CONFIGURING DCC
DCC Configuration 
Methods3Com routers provide two DCC configuration methods: circular DCC, and 
resource-shared DCC. With distinguishing features, these two methods are 
applicable to different applications. In applications, the participating parties of a 
call can flexibly select either method as needed. In other words, one party can 
adopt circular DCC while the other party adopt resource-shared DCC to originate 
a call. 
Circular DCC
Circular DCC has the following features:
■A logical dial (dialer) interface can use the services provided by multiple physical 
interfaces (such as Serial0). However, a physical interface can only belong to 
one dialer interface. That is, a physical interface can only provide one type of 
dial service.
■The user can either bind a physical interface to a dialer interface for inheriting 
the DCC parameters by assigning it to a dialer circular group, or directly 
configure DCC parameters on the physical interface.
■All the physical interfaces served for the same dialer circular group inherit the 
attributes of the same dialer interface.
■Through configuring the dialer route command, a dialer interface can be 
associated with multiple dialing destination addresses. Through configuring the 
dialer number command, however, a dialer can only be associated with one 
dialing destination address.
In addition, all the B channels on an ISDN BRI interface inherit the configuration of 
this physical interface, and the dial route will become more complicated as the 
network grows and more protocols are supported. Therefore, the application of 
circular DCC is restricted due to the static binding between the dialing destination 
addresses and the physical interface configuration. 
Association between the physical interfaces and dialer interfaces in 
circular DCC
Figure 224   Association between the physical interfaces and dialer interfaces in Circular 
DCC
As shown in Figure 224, in the case that dialer interfaces are used, a physical 
interface can only belong to one dialer interface, but each dialer interface can 
Dialer1
Dialer2
Destination A
Destination B
Destination C Physical
interfacesDialer
interfaces
Serial0
Bri1
Serial1
Bri0
Serial2
Async0dialer route
dialer routedialer number