Page 206 / 216 Scroll up to view Page 201 - 205
Administrator’s Handbook
206
Page 207 / 216
207
Appendix B
Quality of Service (QoS) Examples
This section contains information about the Motorola Gateway QoS implementation.
Overview
When packets arrive on a high speed interface and are forwarded to a low speed interface, there is
contention for bandwidth. This is the use case for QoS: to make effective use of bandwidth.
The basic steps for Quality of Service are to match and identify packets as belonging to a class of traf-
fic, and to give each class of traffic a certain behavior such as priority queuing or bandwidth shaping
across critical networking bottlenecks.
Packets forwarded through the system are classified using sets of filter rules to match various criteria,
for example p-bit, DSCP, IP address, port, etc. The matching rule can set the classification, which is
the name of the queue that is to be used.
Figure 1. Illustration of upstream congestion, all traffic is consistently delayed.
Figure 2. Illustration of classification and transmit queue in a simple high/low priority scheme. Low priority may transmit
only when high priority is completely empty.
Page 208 / 216
Administrator’s Handbook
208
Figure 3. Illustration of priority scheduling
Figure 4. Illustration of weighted fair queue scheduling
Figure 5. Illustration of a hybrid queue that is both priority and WFQ, to both constrain bandwidth usage and expedite
one of the queues.
After the packet has been classified, it can be put in the proper queue. Queues are assigned to inter-
faces and can be constructed of several queue components to deliver the desired behavior. The com-
ponents of an interface queue are these building blocks:
basic queues: a 1 input 1 output packet list with a length of 64 packets by default. Packets will tail-
drop when the enqueued to a full basic queue.
priority queue: 2 or more input, 1 output. Schedules the packets from the various inputs strictly
according to input's priority.
weighted fair queue: 1 or more input, 1 output. Schedules the packets according to bandwidth con-
straints
Packets are enqueued to basic queues, and only to basic queues. Basic queues are output to priority
queues and weighted fair queues, which act as “plumbing” elements that alter the dequeuing order
and rate, respectively. Priority queues and weighted fair queues can contain one another.
Page 209 / 216
209
Weighted fair queues are used to constrain bandwidth. For example, consider a weighted fair queue
with three basic queues as inputs, EF, AF and BE:
WFQ
Input 1: EF
Input 2: AF
Input 3: BE
Each input entry is configured with a weight value, which is the rate at which to limit the traffic. This
weight can be either absolute (bps) or a relative percentage of the interface's data-rate. This allows
dedicating a split amount of bandwidth to each queue. A special value for the weight parameter is zero,
which will use the remainder of unclaimed bandwidth.
There is an option to enable bandwidth sharing, so that unused bandwidth in idle queues can be
shared to other queues. When the traffic resumes in the previously idle queue, the previously shared-
out bandwidth is taken back.
When bandwidth sharing is enabled, a secondary rate configuration appears on each input entry, the
peak parameter. This is a hard limit on the amount of bandwidth that the particular input entry can use.
This rate will not be exceeded, even if there is an excess pool of idle bandwidth that could otherwise
be shared.
Upstream QoS: Priority and shaping
The Gateway uses the DSL sync rate to determine traffic shaping requirements for WAN traffic. In this
case there are 6 basic queues, and a hierarchy of both priority queue and weighted fair queue with
bandwidth sharing and dual rate shaping. First the packets are classified via the filterset, to set the
QoS-marker with the name of the desired basic queues. The queues are shown here, with packets
traveling from left to right. Each basic queue feeds into a WFQ entry, and is shaped between the mini-
mum bandwidth defined by “weight”, and the maximum rate defined by “peak”. If there is sufficient
bandwidth, the WFQ entry shapes at the peak rate. If there is no spare bandwidth available for shar-
ing, then the queue is shaped at the “weight” rate. The weight” rate is defined either as a bps value, or
as a percentage of line-rate that is determined once the upstream WAN data-rate is acquired. This
“weight” value behaves as a Committed Information Rate (CIR), and the “peak” value behaves as a
Peak Information Rate (PIR.)
Figure 6. Illustration of default queues used for AT&T
Page 210 / 216
Administrator’s Handbook
210
Packet Rx > Filterset Rules:
Match Rule 1? Set QoS marker = EF
Match Rule 2? Set QoS marker = AF1
Match Rule 3? Set QoS marker = BE
EF
CIR/PIR -> wfq_hi
-> PQ1 \
AF4 CIR/PIR \
> PQ output to interface Tx
AF3 CIR/PIR
\
/
AF2 CIR/PIR
> wfq_lo > PQ2 /
AF1 CIR/PIR
/
BE
CIR/PIR /
Downstream QoS: Ethernet Switch
The simplest way of handling downstream QoS (from WAN to LAN) is to use the per-port queues that
are present in the Ethernet switch. This achieves the greatest efficiency since the queues are handled
in the switch hardware, and should be used when a strict priority queue with 4 priorities is sufficient.
The traffic is classified by priority-bit value. This can be the value retained from WAN ingress (assum-
ing WAN is tagged,) or it can be a value that is set via a filter rule, which allows for advanced classifica-
tion criteria to be used. Even though the LAN interface might not be tagged, there is still an internal
priority field which is used to convey this information to the switch.
Downstream QoS: Egress queues
The secondary method of downstream QoS is to assign egress queues to the LAN port configuration.
This is less efficient, however it allows more advanced queue scheduling algorithms to be used. Pack-
ets are classified by QoS markers set by filter rules.
This method is typically not recommended for deployment configuration as
this mechanism can consume a large amount of CPU processing band-
width.

Rate

4.5 / 5 based on 2 votes.

Bookmark Our Site

Press Ctrl + D to add this site to your favorites!

Share
Top