3Com Router User Manual

							Physical Interface Line Rate677
Shape all the flows on Ethernet interface 1.
[Router] interface ethernet1
[Router-Ethernet1] qos gts any cir 45000000 cbs 5800000 ebs 5800000
Physical Interface Line 
RateBy using the physical interface line rate (LR), the total rate for sending packets 
(including the emergency packet) on a physical interface can be limited.
LR also uses the token bucket to perform the traffic control. If LR is configured in 
an interface of the router, the LR token bucket first processes all the packets sent 
by this interface. If the token bucket has sufficient tokens, the packet can be sent; 
otherwise, the packet enters the QoS queue for congestion management. Thus, 
the packet traffic through this physical interface can be controlled.
Figure 216   Schematic diagram of LR processing
As the token bucket is used to control the traffic, when there is any token in the 
token bucket, the burst transmission of the packet is allowed. When there is no 
token in the token bucket, the packet cannot be sent until a new token is 
generated in the token bucket. Thus, there is a limitation that packet traffic cannot 
be larger than the generating speed of the token, therefore, it realizes that the 
traffic is limited and burst traffic is allowed to pass through at the same time.
Compared with CAR, LR can limit all the packets passing through the physical 
interface. CAR is implemented in the IP layer and is ineffective on the packets that 
are not processed by the IP layer. It is simple to use LR when the user only requires 
the limitation of all packets.
LR ConfigurationTo configure the physical interface line rate, perform the following configurations 
in the interface view.
Ta b l e 717   Configure the Physical Interface LIne Rate
By default, the line rate is not performed on the physical interface.
incoming packetsoutgoing packets
Token Bucket Tokens enter bucket
at the given speedclassify
buffer
queue
OperationCommand
Configure the physical interface 
bandwidthqos lr cir committed-rate [ cbs 
burst-size [ ebs excess-burst-size ] 
]
Delete the configured physical interface 
bandwidthundo qos lr 
						

							678CHAPTER 48: TRAFFIC POLICING, TRAFFIC SHAPING AND LINE RATE
Displaying and 
Debugging LRTa b l e 718   Display and Debug LR
OperationCommand
Display the LR configuration conditions 
and statistic information of the interfacedisplay qos lr [ interface type 
number ] 
						

							49
CONGESTION MANAGEMENT
This chapter covers the following topics:
■What is Congestion?
■Congestion Management Policy Overview
■Selecting Congestion Management Policies
■Operating Principle of the Congestion Management Policies
■Configuring Congestion Management
■Congestion Management Configuration Examples
What is Congestion?For a network unit, when the speed of the data packet is faster than the speed at 
which this interface sends the data packet, congestion occurs on the interface. If 
not enough memory space can be provided to store these data packets, some of 
them will be lost. The loss of the data packet can cause the host or router that is 
sending the data packet to resend this data packet because of a timeout which 
can cause a communication failure.
There are many factors causing congestion. For example, when the data packet 
flow enters the router through the high-speed link and is then transmitted 
through the low speed link, congestion can occur. When the data packet flow 
enters the router simultaneously from multiple interfaces and is transmitted from 
one interface or the processor slows down, congestion may occur.
As shown in Figure 217, two LANs of one company are connected with each other 
through the low speed link. When a user on LAN 1 sends a large number of data 
packets to a user on LAN 2, it may cause congestion on the interface through 
which router A of LAN 1 is connected to the low speed link. If an important 
application is running between the servers of both LANs, while an unimportant 
application is running between two PCs, the important application will be 
influenced. 
						

							680CHAPTER 49: CONGESTION MANAGEMENT
Figure 217   Schematic diagram of the congested network
Congestion 
Management Policy 
OverviewWhen the congestion occurs, if not enough memory space is provided to buffer 
the packets, some of the packets will be lost. The loss of the packets may cause 
the host or router that is sending the packet to resend this packet because of 
overtime, re-congesting and resending, and so on, thereby causing a vicious circle. 
Therefore, some policies are used to manage network congestion. When 
congestion occurs, the router takes some policies to dispatch the data packets, 
deciding which data packets may be sent first and which ones may be discarded. 
These policies are called the congestion management policy.
For the congestion management, the queuing mechanism is generally used. When 
congestion occurs, the packet is queued at the router egress by a given policy. 
During dispatching, the order for sending the packet out of the queue is decided 
by a given policy.
FIFO QueuingIn the FIFO mode, the concept of no communication priority and classification is 
adopted. During the use of FIFO, the sending order of data packet from the 
interface depends on the order in which the data packet arrives at this interface, at 
this time, the queuing and de-queuing orders of the packet are the same.
FIFO provides the basic storage and transmission capabilities. 
Priority QueuingIn Priority Queueing (PQ) mode, you can flexibly specify the priority queues which 
the packets enter according to the fields packet length, source address, and 
destination address in the packets header and the interface into which the packets 
will come. The packets belonging to a higher priority queue can be sent first. In 
this way, the most important data can be handled first.
Custom QueuingIn the Custom Queueing (CU) model, according to the users requirements, the 
traffic can be classified in terms of TCP/UDP port number, ACL and interface type. 
Each type of traffic is allocated with a certain percent of bandwidth. When 
network congestion occurs, the traffic that has high demands on delay (such as 
voice) can obtain reliable service. If a type of traffic cannot occupy all the reserved 
bandwidth, other types of traffic will occupy the reserved bandwidth 
automatically, thus making full use of the resource.
DDN/FR/ISDN/PSTN
Quidway
Router
Ethernet
Server
Ethernet
Server
Company LAN1
10 M
100 M
Occurrence of
congestion
P
C
P
C
Company LAN2
RouterARouterB 
						

							Selecting Congestion Management Policies681
For the interface with the lower rate, customizing the queue for it can guarantee 
that the data flows passing through this interface may also obtain the network 
services to certain extent.
Weighted Fair QueuingWeighted Fair Queuing (WFQ) provides a dynamic and fair queuing mode, which 
distinguishes the traffic based on the priority/weight and decides the bandwidth 
size of each session according to the session situation. Thus, it guarantees that all 
communications can be fairly treated according to the weight allocated to them. 
The foundation based on which WFQ classifies the traffic includes the source 
address, destination address, source port number, destination port number, and 
protocol type.
Selecting Congestion 
Management Policies3Com routers implement the four congestion management policies (FIFO, PQ, CQ 
and WFQ) discussed previously, in the Ethernet interface and serial interface 
(encapsulated PPP, FR, HDLC), which may satisfy the requirements for various 
service qualities to a certain extent.
FIFO implements the no priority policy of the data packet in user data 
communication, which is not needed to determine the priority or type of the 
communication. However, when using the FIFO policy, some low priority data in 
abnormal operation may consume most of available bandwidths and occupy the 
entire queue, which causes the delay of the burst data source, and the important 
communication may be thereby discarded.
PQ can assure some communication transmission with higher priority. That is, the 
strict priority sequence is conducted at the cost of transmission failure of data 
packets with lower priority. For example, the packets in the lower priority queue 
may not be transmitted in the worst case where the available bandwidth is very 
limited and emergency communication occurs frequently.
CQ reserves a certain percent of available bandwidth for each type of specified 
traffic, so that the interface running at a low rate can obtain network service even 
if congestion occurs. The size of this queue is determined by deciding the total 
number of the data packets configured in the queue to control access to the 
bandwidth.
WFQ uses the fair queuing algorithm to dynamically divide the communications 
into messages. The message is a part of a session. With the use of WFQ, the 
interactive communication with a small capacity can obtain the fair allocation of 
the bandwidth, as the same as the communication with a large capacity (such as 
file transmission).
Ta b l e 719 compares between the four different policies: 
						

							682CHAPTER 49: CONGESTION MANAGEMENT
Ta b l e 719   Comparison of Several Congestion Management Policies
Operating Principle of 
the Congestion 
Management PoliciesFor congestion management, queuing technology is used. When congestion 
occurs, the data packet is queued at the router by a policy. When dispatching, the 
order for sending the data packet is decided by the policy.
Number of 
queues
AdvantageDisadvantage
FIFO11. It does not need to be 
configured and is easy to use.
2. The processing is simple with 
small delay.
1. No matter how urgent they are, 
all the packets, voice or data, will 
enter the FIFO (First In, First Out) 
queue. The bandwidth used for 
sending packets, delay time, drop 
rate are decided by the arrival 
sequence of the packets.
2. It has no restriction on the 
uncoordinated data sources (such as 
the packet transmission of UDP), and 
the unmatched data sources will 
cause the damage of the 
coordinated data source bandwidth 
(such as the TCP packet 
transmission).
3. The delay of the real time 
application sensitive to time (such as 
VolP) cannot be guaranteed.
PQ4The absolute priority can be 
provided to various service data, 
and the delay of the real time 
application sensitive to time 
(such as VolP) can be 
guaranteed. The bandwidth 
occupation of the packet with 
the priority service may have the 
absolute priority.1. It needs to be configured, and the 
processing speed is slow.
2. If the bandwidth of the packet 
with high priority is not restricted, it 
will cause that the packet with low 
priority cannot obtain the 
bandwidth.
CQ11. The packets of various services 
may be allocated with the 
bandwidths based on the 
bandwidth proportion.
2. When there is no packet, the 
available bandwidth occupied by 
the existing types of packets can 
be automatically increased.
It needs to be configured, and the 
processing speed is slow.
WFQIt is decided 
by users
(256 by 
default)
1. It is easily configured.
2. The bandwidth of the 
coordinated (interactive) data 
source (such as the TCP packet 
transmission) can be protected.
3. The delayed jitter can be 
reduced.
4. The small packet has priority.
5. The flows with various priority 
levels may be allocated with 
different bandwidths.
6. When the traffic is reduced, 
the available bandwidth 
occupied by the existing flows 
may be automatically increased.
The processing speed is slower than 
FIFO. 
						

							Operating Principle of the Congestion Management Policies683
Figure 218   Schematic diagram of the first in first out queue
First-In, First-Out (FIFO) 
QueuingAs shown in Figure 218, the data packets are input to the first-in, first-out (FIFO) 
queue according to the priority order of their arrivals. Data packets that first arrive 
are first transmitted, and the data packets that later arrive are transmitted later. All 
the packets that will be transmitted from the interface are input to the end of the 
FIFO queue of the interface in the priority order of their arrivals. At the time when 
the interface transmits the packets, the packets are transmitted in order, starting 
from the head of the FIFO queue. During the transmission process of all packets, 
there is no difference and no guarantee is provided for the quality of the packet 
transmission. Therefore, a single application can occupy all the network resources, 
seriously affecting the transmission of key service data.
Priority Queuing (PQ)As shown in Figure 219, the PQ queue is used to provide strict priority levels for 
important network data. It can flexibly specify the priority order according to the 
network protocol (such as IP or IPX), the interface into which the data are input, 
the length of the packet, and the source address, destination address, and other 
features.
Figure 219   Schematic diagram of the priority queuing
When the packets arrive at the interface, all of them are first classified (up to 4 
classifications), and then they are input to the ends of respective queues according 
to the classifications of the packets. Upon the transmission of the packets, 
according to different priority levels, the packets in the low priority queue are not 
transmitted until all the packets in the high priority queues are transmitted. Thus, 
it is guaranteed that, at the network unit where the PQ is utilized, the most 
important data can be processed the soonest and the packets of the higher 
priority queues have very low delay. Both packet performance exponents of loss 
incomi ng pack et s
queue
out goi ng pac k et s
queuei nginterface
incomi ng pack et s
t op queue
mi ddl e queue
cl assi f yi ng
out goi ng pack et s
nor mal  queue
bot t om queue
queuei nginterface 
						

							684CHAPTER 49: CONGESTION MANAGEMENT
rate and throughput rate can be guaranteed to a certain extent in case of network 
congestion.
The key service (such as ERP) data packets may be put into the higher priority 
queue, while the non-key service (such as E-Mail) data packets are put into the 
lower priority queue, so that the data packets of the non-key service are 
transmitted in the idle intervals during the processing of the key service data. In 
this way, the priority of the key service is guaranteed and network resources are 
optimized. However, it brings the problem that the data packets in the lower 
priority queue may be blocked in the packet queue of the transmission interface 
for a long period because of the existence of the data packets in the higher 
priority queue.
Custom Queuing (CQ)As shown in Figure 220, custom queuing (CQ) divides the data packets into 17 
classifications (corresponding to 17 queues of CQ) according a given policy, and 
data packets are input respective CQ queues based on their own classifications 
following the FIFO policy. In 17 queues of CQ, the queue 0 is the system queue, 
and queues 1 to 16 are the user queues. The users can configure the proportional 
relationship of the occupied interface bandwidth between various user queues. 
When dispatching the queue, the data packets in the system queue are first 
transmitted. Before the system queue is empty, a certain number of data packets 
from user queues 1 to 16 are not extracted and sent out according to the 
predetermined configured proportion using polling method. 
Figure 220   Schematic diagram of the custom queuing
PQ assigns the absolute priority to the data packets with higher priority compared 
to data packets with the lower priority level. In this way, though the priority 
transmission of the key service data can be guaranteed, when a number of data 
packets with higher priority need to be transmitted, all bandwidths may be 
occupied, causing the data packets with lower priority to be completely blocked. 
With the use of CQ, such a case can be avoided. CQ has total of 7 queues. Queue 
0 is the system queue that is first dispatched, and the queues 1 to 16 are the user 
queues that are dispatched by a polling method based on the bandwidth settings. 
The users may configure the proportional relationship of the occupied bandwidth 
between the queues and the enqueuing policy of the packets. Thus, the data 
packets of various services can be provided with different bandwidths, to 
guarantee that the key services can be provided with more bandwidth. In addition, 
it is not likely that non-key services may not be allocated with the bandwidth.
incomi ng packet s
queue1
queue2
cl assi f yi ng
out goi ng pack et s
queue15
queue16
queuei nginterface
¡ -¡-10%
30%
10%
5% 
						

							Operating Principle of the Congestion Management Policies685
In the network shown in Figure 217, it is assumed that the server of LAN 1 
transmits the data of the key service to the server of LAN 2, and the PC of LAN 1 
transmits the data of the non-key service to PC of LAN 2. If the serial interface to 
be connected with the WAN is configured for congestion management with CQ, 
and the data flows of the key services between the servers are input to queue A, 
while the data flows of the non-key services are input to queue B, the proportional 
relationship of the occupied interface bandwidth between queue A and queue B is 
configured as 3:1 (for example, during dispatching, queue A may continuously 
transmit 6000 bytes of data packets every time, while queue B may continuously 
transmit 2000 bytes of data packets every time). Thus, CQ will treat the data 
packets of both different services differently. Each time queue A is dispatched, the 
data packets are continuously transmitted, before the transmitted bytes are not 
less than 6000 or queue A is empty, the next user queue will not be dispatched. 
When queue B is dispatched, the condition to stop dispatching is that the 
continuously transmitted bytes are not less than 2000 or queue B is empty. 
Therefore, when congestion occurs and there are data packets in queues A and B 
ready to be transmitted, in the view of the statistic results, the proportion between 
the bandwidths allocated to the key services and the bandwidths allocated to the 
non-key services is approximately 3:1.
Weighted Fair Queuing 
(WFQ)Weighted fair queuing (WFQ), is based on the guarantee of fair bandwidth delay, 
and reflects the weighted value that is dependent on the PI priority carried in the 
IP packet header. As shown in 
Figure 221, weighted fair queuing classifies the 
packets based on the flows (identical source IP address, destination IP address, 
source port number, destination port number, protocol number, and ToS packets 
that belong to the same flow), with each flow allocated to one queue. When 
dequeuing, WFQ allocates the available bandwidth of the egress to each flow. The 
smaller the value of the priority is, the less the allocated bandwidth is. The larger 
the value of the priority is, the more the allocated bandwidth is.
Figure 221   Schematic diagram of weighted fair queuing
The occupied bandwidth proportion of each flow is: 
(its own priority level+1)/(the sum of all of them (the priority levels of the flows+1))
For example, there are 5 types of traffic on an interface, and their priority levels are 
0,1,2,3 and 4 respectively, the total quota of the bandwidth is the sum of each 
priority plus 1, that is 1 + 2 + 3 + 4 + 5 = 15. The percentage of the bandwidth 
incomi ng pack et s
queue1 wei ght 1
queue2 wei ght 2
cl assi f yi ng
out goi ng pack et s
queueN- 1 wei ght N- 1
queueN wei ght N
queuei nginterface
¡ -¡- 
						

							686CHAPTER 49: CONGESTION MANAGEMENT
occupied by each traffic is (each priority + 1)/ the sum of each priority plus 1, that 
is, 1/15, 2/15, 3/15, 4/15 and 5/15.
For example, there are total 4 flows currently, and the priority levels of three of 
them are 4, and that of one of them is 5, and then the total number of the 
allocated bandwidth is:
(4 + 1) x 3 + (5 + 1) = 21
Then, the bandwidths of the three flows with the priority levels of 4 are 5/21, and 
the bandwidth of the flows with the priority level of 5 is 6/21.
Configuring 
Congestion 
ManagementThis section describes the following types of congestion management:
■Configuring FIFO Queuing
■Configuring Priority Queuing
■Configuring Custom Queuing (CQ)
■Configuring WFQ
Configuring FIFO
QueuingTo configure FIFO queuing, perform the following configurations in the interface 
view.
Ta b l e 720   Configure the First In First Out Queuing
By default, the length of the FIFO queue is 75, with the value ranging 1 to 1024.
Configuring Priority 
QueuingPriority queuing configuration includes:
■Configuring priority queuing
■Applying the priority-list queuing group to the interface
■Specifying the queue length of the priority-list queuing 
Configuring priority queuing
The priority queuing classifies the packets according to a given policy, and all the 
packets are divided into 4 classifications, each of which corresponds to one of the 
4 queues of PQ respectively. Then the packet is input to the corresponding queue 
according to its classification. The 4 PQ queues are: high priority queue (top), 
medium priority queue (middle), normal priority queue (normal) and low priority 
queue (bottom) with priority levels decreased sequentially. Upon the transmission 
of packets, they are sequentially transmitted according to their priority orders, that 
is, the packets in the top queue are first transmitted, then the packets in the 
middle queue are transmitted, and then the packets in the normal queue are 
transmitted, and finally the packets in the bottom queue are transmitted.
The priority queuing includes up to 16 groups (the value range of pql-index is 1 to 
16), each of which specifies which types of data packets input which queue, the 
OperationCommand
Configure the length of FIFO queueqos fifo queue-length queue-length
Recover the default value of the FIFO 
queue lengthundo qos fifo queue-length