1. D-Link QoS Technology Brief
Arthur Chiang
Ethernet Switch Product Dept.
Jan 2011 HQ

2. Agenda What is QoS Policing Queuing Prioritizing Rate Limiting
Three Color Marker
Queuing
Scheduling
Congestion Control
Shaping

3. QoS What is Quality of Service (QoS) Definition Policing Queuing
QoS is a set of techniques to manage network resources, including delay, delay variation, bandwidth and packet loss.
Users or data flows are able to be guaranteed a certain level of performance through these functions.
Policing
QoS
Queuing

4. What is Quality of Service (QoS) Explanation of Features
Policing
A given policy to prioritize or limit the maximum data rate of traffic.
Policy can be port-based, flow-based, queue-based.
Queuing
Manages the data stored in device buffer memory.
Decides the sequence of data forwarding.
Allocates the buffer resource. 4

5. Agenda What is QoS Policing Queuing Prioritizing Rate Limiting
Three Color Marker
Queuing
Scheduling
Congestion Control
Shaping 5

6. Prioritizing Priority Tags
Two kinds of priority tags
IEEE 802.1p tag in L2 Ethernet header
Usually inserted by switches as part of 802.1Q VLAN tag according to the policy bound on the ingress port.
Might not be carried across L3 networks as it is a L2 tag (Note)
DSCP (IP Precedence) tag in L3 IP header.
A field of IP header with default value 0.
Usually modified by applications such as VoIP, IP-TV, and P2P based on the prioritization needs.
Can be carried across L3 networks
Data
L3 Header
ToS
802.1Q
L2 Header
VLAN ID
CFI
802.1p
3bits
1bit
12bits
Delay/Cost/…
5bits
DSCP
8bits
IP Precedence
Routers will remove L2 header when doing L3 routing process. When routing is done, the 802.1p tag will be added along with the 802.1Q tag before delivering. Routers will decide the new 802.1p value according to the configured priority-queue map. That is, if routers change the priority queue mapping or use L3 priority for queue mapping, the L2 priority will be changed.
Differentiated Services Code Point (DSCP)
The original definition of Type of Service field including IP Precedence, Delay, Throughput, Reliability, Cost. All these fields were combined in RFC 2474 for Differentiated services (DiffServ). 6

7. Prioritizing Sequence of Packet Forwarding
Priority Queues
Switch will store all the packets into port egress buffer before transmitting it.
Each egress buffer is divided into several priority queues and packets stored in higher queues will be served first.
Generally speaking, packets with higher priority tag will be mapped to higher queues. However, it can be changed according to the policies of switch itself.
Add Priority Tag
Queue Mapping
Ingress
Egress
Queue Buffer
Highest Priority
Higher Priority
Lower Priority
Lowest Priority 7

8. Policy (Default Port Priority) Policy (Default Port Priority)
Prioritizing Applications of Prioritizing
Prioritize traffic of endpoints on different ports
Insert the priority tag to untagged packets according to the port default priority.
Map packets with higher priority tag to higher queue.
Policy (Default Port Priority)
Action: Add 802.1p tag with low priority
Policy (Default Port Priority)
Action: Add 802.1p tag with middle priority
High Priority
Middle Priority
Low Priority
No Priority
Client-2
IP Phone
Client-1 8

9. Advanced Prioritizing Change Priority Tag and Queue Mapping
Using ACL to match specific flow (eg. IP, protocol type) and force the change of priority tag of packets.
Change queue mapping
Using either L2 (802.1p) or L3 (DSCP) priority tag to map the system queue.
Force the change of the queue mapping for specific flow by using ACL.
Change Priority Tag
Change Queue Mapping
Add Priority Tag
Queue Mapping
Change Priority Tag
Ingress
Egress
Queue Buffer
Highest Priority
Higher Priority
Lower Priority
Lowest Priority 9

10. Advanced Prioritizing Applications of Prioritizing
Prioritizes different network services on the same port
If two or more applications are in the same port, replace the priority tag of critical applications to make sure it will be served first.
Local prioritization
For a user or application that need priority in local network only: Changing the priority queue mapping, instead of replacing the priority tag, can limit the impact of the prioritization.
Same service level in the internet
Policy (ACL Rule)
Match: VoIP Protocol
Action: Replace priority tag with ‘high’
Client-2
Policy (ACL Rule)
Match: Client-1 MAC Address
Action: Force mapping to high queue
ERP Server
Client-1
High Priority
Middle Priority
Low Priority
Client-2
IP Phone
Client-1 10

11. Summary Planning Priority Policy for Network
Guarantee the service level in the network
Set priority policies on access ports of edge switches.
Keep the consistent queue mapping policy among the network to save system resource of backbone switches.
Policy: Voice changes to high priority
High priority maps to high queue
High priority maps to high queue
High priority maps to high queue
High priority maps to high queue
Voice
Policy: Data changes to low priority
Low priority maps to low queue
Low priority maps to low queue
Low priority maps to low queue
Low priority maps to low queue
Data
High Priority
Low Priority 11

12. Rate Limiting Definition of Rate Limiting
A given threshold for data flow. User can set up QoS actions for traffic exceeding the threshold. Such as: - Drop (Bandwidth Control) - Replace priority tag - Delay forwarding (Shaping)
bps
Action
Rate
Time
Rate Limiting 12

13. Rate Limiting Type of Rate Limiting
Port-based Rate limiting
Hard limits the ingress/egress data rate per physical port, regardless of the content of the data.
Manages the data rate at access layer devices, preventing the overloading of backbone networks.
Flow-based Rate limiting
Controls the data rate of a specific flow by using ACL rules, for example, a specific IP or L4 protocol.
Well manages the bandwidth or QoS policies for service running on the physical ports.
Queue-based Rate limiting
Hard limits the ingress/egress data rate per egress queue basis, regardless of the content of the data.
Prevents the egress bandwidth from being occupied by a specific queue, especially in the applications of Strict Priority scheduling and 3 color marker.
Note: Most of the switch controllers support only drop action on Queue-based Rate Limiting nowadays 13

14. Rate Limiting Advanced Rate Limiting
For some critical applications or customer-signed Service Level Agreement (SLA), administrators define two thresholds for better bandwidth allocation and service quality.
Committed Information Rate (CIR): A guaranteed data rate of traffic. The sum of all CIR must be smaller than corresponding physical interface or the bandwidth cannot be guaranteed.
Peak Information Rate (PIR): A maximum data rate of traffic. Sometimes called Exceed Information Rate (EIR). Usually, switch will do best-effort delivery for traffic exceeding CIR and drop the traffic exceeding PIR. However, the action can be changed depending on different environment.
bps
Action
bps
Action 2
Rate
PIR
Action 1
CIR
Time
Single Rate
Time
Two Rate 14

15. Rate Limiting Advanced Rate Limiting – Three Color Marker
Three Color Marker borrows the DSCP value in IP header as the color code. There are three color codes – red, yellow and green, and the ‘DSCP’ to ‘color code’ mapping can be defined by users.
Two different Color Marker mechanisms:
RFC2697 Single Rate Three Color Marker (srTCM): Adds color tag according to configured buffer size; Focuses more on the size of packets.
RFC2698 Two Rate Three Color Marker (trTCM): Adds color tag according to data rates; Focuses more on the data rates.
Two different operating modes for 3 color marker
Color blind mode: Do not trust the original DSCP value in packet; assume all packets are uncolored. Usually used at access layer network.
Color aware mode: Trust the original DSCP value in packet; assume all packets are colored. Usually used at aggregation/core layer. 15

16. Three Color Marker Single Rate Three Color Marker
The srTCM meters a traffic stream and marks its packets according to three traffic parameters: CIR, Committed Buffer Size (CBS) and Excess Buffer Size (EBS).
A packet is marked green if it does not exceed the CBS, yellow if it does exceed the CBS, but not the EBS, and red otherwise.
Buffer Memory
Mark Green
Mark Red
CIR
The behavior of the Meter is specified in terms of its mode and two token buckets, C and E, which both share the common rate CIR. The maximum size of the token bucket C is CBS and the maximum size of the token bucket E is EBS.
The token buckets C and E are initially (at time 0) full, i.e., the token count Tc(0) = CBS and the token count Te(0) = EBS. Thereafter, the token counts Tc and Te are updated CIR times per second as follows:
o If Tc is less than CBS, Tc is incremented by one, else
o if Te is less then EBS, Te is incremented by one, else
o neither Tc nor Te is incremented.
When a packet of size B bytes arrives at time t, the following happens if the srTCM is configured to operate in the Color-Blind mode:
o If Tc(t)-B >= 0, the packet is green and Tc is decremented by B down to the minimum value of 0, else
o if Te(t)-B >= 0, the packets is yellow and Te is decremented by B down to the minimum value of 0, else
o the packet is red and neither Tc nor Te is decremented.
When a packet of size B bytes arrives at time t, the following happens if the srTCM is configured to operate in the Color-Aware mode:
o If the packet has been precolored as green and Tc(t)-B >= 0, the packet is green and Tc is decremented by B down to the minimum value of 0, else
o If the packet has been precolored as green or yellow and if Te(t)-B >= 0, the packets is yellow and Te is decremented by B down to the minimum value of 0, else
Mark Yellow
CBS
EBS 16

17. Three Color Marker Two Rate Three Color Marker
The trTCM meters a traffic stream and marks its packets according to two traffic parameters: CIR and PIR.
A packet is marked green if it does not exceed the CIR, yellow if it does exceed the CIR, but not the PIR, and red if it exceed PIR.
Buffer Memory
Mark Red
PIR
Mark Yellow
CIR
Mark Green
The token buckets P and C are initially (at time 0) full, i.e., the token count Tp(0) = PBS and the token count Tc(0) = CBS. Thereafter, the token count Tp is incremented by one PIR times per second up to PBS and the token count Tc is incremented by one CIR times per second up to CBS.
When a packet of size B bytes arrives at time t, the following happens if the trTCM is configured to operate in the Color-Blind mode:
o If Tp(t)-B < 0, the packet is red, else
o if Tc(t)-B < 0, the packet is yellow and Tp is decremented by B, else
o the packet is green and both Tp and Tc are decremented by B.
When a packet of size B bytes arrives at time t, the following happens if the trTCM is configured to operate in the Color-Aware mode:
o If the packet has been precolored as red or if Tp(t)-B < 0, the packet is red, else
o if the packet has been precolored as yellow or if Tc(t)-B < 0, the packet is yellow and Tp is decremented by B, else 17

18. Three Color Marker Color Blind/Aware Mode Process
Color Blind Mode
Meters and marks color tag to the packet according to the CIR/PIR or CBS/EBS.
Drop Red packets and remark the DSCP tag to Green or Yellow according to the color tag.
Color Aware Mode
Meters the Green and Yellow traffic and mark to Yellow or Red according to the CIR/PIR or CBS/EBS.
Drop Red packets and remark the DSCP tag of Green packets to Yellow for over-threshold Green traffic.
Revise Color Tag
3 Color Marker
Drop Policer
Queue Mapping G R Y R Y R Y G
Ingress
Egress
Queue Buffer
Queue Buffer 18

19. Physical port bandwidth
Three Color Marker Planning a Service Guaranteed Network with Three Color Marker
Set color blind mode and high CIR for important services at access switches.
Set color aware mode to simplify the configuration and unify the QoS policy on aggregation switches.
Make sure the green traffic won’t exceed the physical bandwidth.
Policy: - VoIP, Admin with high CIR, PIR - Employee with middle CIR, PIR - Guest with low CIR, PIR
Action:
- Remark DSCP of Green/Yellow packet
- Drop Red packet
Policy: - Green: Transparent (Do nothing) - Yellow: Set PIR for yellow
Action:
- Drop Red packet …
Physical port bandwidth …
Green Traffic
Traffic 19

20. Rate Limiting Summary
For single user/service per port, use port-based rate limiting with drop action to control the traffic bandwidth.
For multiple users/services per port, use flow-based rate limiting with drop action to prevent the interference between each service.
To maximize the port bandwidth usage and also guarantee the service level of each service in a multiple users/services per port, use three Color marker function to fine tune the traffic among the port.
Mark Green for minimum bandwidth of each service to make sure it will be served first.
Mark Yellow for maximum bandwidth of each service, doing best effort forwarding when traffic conflict.
Mark Red for traffic exceed the maximum bandwidth of each service, drop these packets to limit the maximum bandwidth of this port.
IPTV
VoIP
CIR
CIR
PIR
PIR
PIR
2 rate, 3 Color marking (TrTCM) - Typical configuration:
- Green – Forwarded frames – CIR conforming traffic
- Yellow – Discard Eligible frames – Over CIR , within EIR
- Red – Discarded frames – Exceeds EIR
Data
Total UNI (User Network Interface) Bandwidth
CIR

21. Agenda What is QoS Policing Queuing Prioritizing Rate Limiting
Three Color Marker
Queuing
Scheduling
Congestion Control
Shaping 21

22. Scheduling Scheduled Data Forwarding
Traditional scheduling mechanism
First-In First-Out (FIFO) Queuing, Strict Priority Queuing (SPQ), Round-Robin, Weighted Round-Robin (WRR) Queuing
More accurate scheduling is required to make the egress bandwidth usage more properly
Fair Queue (FQ), WFQ, Deficit Round Robin (DRR), WDRR: Forwarding/scheduling decision is made by byte count and utilization of queues, the true weight of each queue.
Smooth Forwarding: Fairly distributes packets with different priorities in each weighted round, making the weight of queues not impacted by priority.
Compound scheduling mechanism is required to well control various IP services nowadays
ST + *RR: Mission critical services, such as VoIP & VOD, have to be forwarded without any delay, while other services can share the rest of bandwidth by using any kind of round-robin mechanism.
Shaped/Shared Round Robin (SRR): SRR is Cisco proprietary scheduling mechanism, it supports following features - Traffic Shaped: Shapes the traffic of each egress queue, including bandwidth control and forward delay. - Shared Bandwidth: Queues can take unused bandwidth when loading is not full. The shared round-robin co-work with Dynamic Buffer Management (DBM) function to automatically fine tune the buffer allocation of each queue. - Smooth Forwarding: Same as D-Link. 22

23. Scheduling Example of Scheduling
High Priority
Middle Priority
Low Priority
FIFO: Do not change the sequence of traffic. First-in, first-out.
SPQ: Packets with higher priority will always be served first. However, if higher queues are always occupied, traffic in lower queues will never be forwarded.
WRR: Packets will be forwarded depending on the assigned weight (number of packets) of each queue. The forwarding ticket will jump to next queue when “weight” is reached or queue is empty. It solves the problem of SPQ, however, the switch’s real output may not match the weight settings if the packet size of each priority is different.
FIFO
SPQ
WRR 23

24. Scheduling Example of Scheduling (continue)
High Priority
Middle Priority
Low Priority
DRR/WFQ: Use byte count instead of packet count as the weight index. The allowed forwarding byte of each queue will increase every round. If the size of next packet is greater than allowed forwarding byte, round-robin mechanism will skip this queue until its allowed forwarding byte is greater than the packet size. It makes the real output more like the weight configuration.
Smooth Forwarding: Makes rapid passes for each queue and use more turns for high queues instead of waiting all packets transmitted in high queues.
DRR/
WFQ
Smoothed
Round
Robin
3rd
2nd
1st round 24

25. Congestion Control Traditional Congestion Control
When ports are overloading, traffic congestion will happen
In traditional device, packets will be dropped when corresponding queue is full (Tail-drop).
However, Tail-drop causes TCP global synchronization (Note) problem when burst traffic happens.
Queue 1
Drop
Queue 2
Ingress
Egress
TCP global synchronization in Computer networks can happen to TCP/IP flows during periods of congestion because each sender will reduce their transmission rate at the same time when packet loss occurs.
Routers on the Internet normally have packet queues, to allow them to hold packets when the network is busy, rather than discarding them.
Because routers have limited resources, the size of these queues is also limited. The simplest technique to limit queue size is known as tail drop. The queue is allowed to fill to its maximum size, and then any new packets are simply discarded, until there is space in the queue again.
This causes problems when used on TCP/IP routers handling multiple TCP streams, especially when bursty traffic is present. While the network is stable, the queue is constantly full, and there are no problems except that the full queue results in high latency. However, the introduction of a sudden burst of traffic may cause large numbers of established, steady streams to lose packets simultaneously.
TCP has automatic recovery from dropped packets, which it interprets as congestion on the network (which is usually correct). The sender reduces its sending rate for a certain amount of time, and then tries to find out if the network is no longer congested by increasing the rate again subject to a ramp-up. This is known as the slow-start algorithm.
Almost all the senders will use the same time delay before increasing their rates. When these delays expire, at the same time, all the senders will send additional packets, the router queue will again overflow, more packets will be dropped, the senders will all back off for a fixed delay... ad infinitum.
This pattern of each sender decreasing and increasing transmission rates at the same time as other senders is referred to as "global synchronization" and leads to inefficient use of bandwidth, due to the large numbers of dropped packets, which must be retransmitted.
Queue 3
Queue 4 25

26. Congestion Control Random Early Detection (RED)
To avoid TCP global synchronization, some new drop algorithm were presented:
Random Early Detection (RED): Drop packets based on the utilization of queues. By randomly dropping packets prior to congestion, RED informs packet sources decreasing its transmission rate. It prevents the TCP global synchronization (Note) problem and makes queue usage fairer when traffic burst happens.
Weighted Random Early Detection (WRED): Associates with the weight of queues, higher drop rate for lower priority queue.
Simple Random Early Detection (SRED): Define the drop rates for color tags in specific queue. Assure the QoS not to be impacted especially for Green packets.
Queue 1
Random Drop
Queue 2
Ingress
Egress
Queue 3
Queue 4
Random Drop
RED Threshold 26

27. Max. Switching Capacity
Shaping Traditional Devices
To lower the packet lose rate, almost all switch controllers have built-in expensive buffer memory to prevent the data re-transmission.
Switch stores the traffic in buffer memory and delays the transmission until the egress bandwidth is available.
However, the best-effort traffic shaping is no longer enough for the QoS requirement nowadays.
Buffer Memory
bps
Store to buffer
bps
Max. Switching Capacity
Time
Time
Data
Data
Delay Forwarding
Delay Forwarding 27

28. Max. Switching Capacity
Shaping Flow-based Traffic Shaping
Due to device buffer memory is limited, reserve the resource for important applications can improve the quality of service.
Configurable buffer size:
Committed Burst Size (CBS) for Green packets
Peak Burst Size (PBS), Excess Burst Size (EBS) for Yellow packets.
Drop
No reservation for
Data exceeding CIR
PIR
CBS for Data
Buffer Memory
PBS for IPTV
bps
CIR
Drop
Drop
Max. Switching Capacity
Data
Store to PBS
Store to PBS
PIR
Time
Store to CBS
Store to CBS
Overall Output
CIR
Delay Forwarding
Delay Forwarding
CBS for IPTV
IPTV 28