Forwarding Classes

Routers support multiple forwarding classes and class-based queuing, so the concept of forwarding classes is common to all of the QoS policies.

Each forwarding class (also called Class of Service (CoS)) is important only in relation to the other forwarding classes. A forwarding class provides network elements a method to weigh the relative importance of one packet over another in a different forwarding class.

Queues are created for a specific forwarding class to determine the manner in which the queue output is scheduled into the switch fabric. The forwarding class of the packet, along with the profile state, determines how the packet is queued and handled (the per hop behavior (PHB)) at each hop along its path to a destination egress point. Routers support eight (8) forwarding classes (Table 2).

Table 2 presents the default definitions for the forwarding classes. The forwarding class behavior, in terms of ingress marking interpretation and egress marking, can be changed by a Network QoS Policies. All forwarding class queues support the concept of in-profile and out-of-profile, and exceed-profile at egress.

The forwarding classes can be classified into three class types:

Queue ID

The queue ID is used to uniquely identify the queue. The queue ID is only unique within the context of the QoS policy within which the queue is defined.

Unicast or Multipoint Queue

Currently, for the 7750 SR and 7950 XRS, only VPLS services use multipoint ingress queues, although IES and VPRN services use multipoint ingress queues for multicast traffic alone when PIM is enabled on the service interface.

Queue Hardware Scheduler

The hardware scheduler for a queue dictates how it will be scheduled relative to other queues at the hardware level. When a queue is defined in a service ingress or service egress QoS policy, it is possible to explicitly define the hardware scheduler to use for the queue when it is applied to a SAP.

Being able to define a hardware scheduler is important as a single queue allows support for multiple forwarding classes. The default behavior is to automatically choose the expedited or non-expedited nature of the queue based on the forwarding classes mapped to it. As long as all forwarding classes mapped to the queue are expedited (nc, ef, h1 or h2), the queue will be treated as an expedited queue by the hardware schedulers. When any non-expedited forwarding classes are mapped to the queue (be, af, l1 or l2), the queue will be treated as best effort by the hardware schedulers.

The expedited hardware schedulers are used to enforce expedited access to internal switch fabric destinations.

Committed Information Rate

The committed information rate (CIR) for a queue performs two distinct functions:

All router queues support the concept of in-profile and out-of-profile. The network QoS policy applied at network egress determines how or if the profile state is marked in packets transmitted into the service core network. If the profile state is marked in the service core packets, out-of-profile packets are preferentially dropped over in-profile packets at congestion points in the core.

Although the router is flexible in how the CIR can be configured, there are conventional ranges for the CIR based on the forwarding class of a queue. A service ingress queue associated with the high-priority class normally has the CIR threshold equal to the PIR rate although the router allows the CIR to be provisioned to any rate below the PIR should this behavior be required. If the service egress queue is associated with a best-effort class, the CIR threshold is normally set to zero; again the setting of this parameter is flexible.

The CIR for a service queue is provisioned on ingress and egress service queues within service ingress QoS policies and service egress QoS policies, respectively.

The CIR for network queues are defined within network queue policies based on the forwarding class. The CIR for the queues for the forwarding class are defined as a percentage of the network interface bandwidth.

Peak Information Rate

The peak information rate (PIR) defines the maximum rate at which packets are allowed to exit the queue. It does not specify the maximum rate at which packets may enter the queue; this is governed by the queue's ability to absorb bursts and is defined by its maximum burst size (MBS).

The actual transmission rate of a service queue depends on more than just its PIR. Each queue is competing for transmission bandwidth with other queues. Each queue's PIR, CIR and the relative importance of the scheduler serving the queue all combine to affect a queue's ability to transmit packets as discussed in Single Tier Scheduling.

The PIR is provisioned on ingress and egress service queues within service ingress QoS policies and service egress QoS policies, respectively.

The PIR for network queues are defined within network queue policies based on the forwarding class. The PIR for the queues for the forwarding class are defined as a percentage of the network interface bandwidth.

When defining the PIR for a queue, the value specified is the administrative PIR for the queue.The router has a number of native rates in hardware that it uses to determine the operational PIR for the queue. The user has some control over how the administrative PIR is converted to an operational PIR should the hardware not support the exact CIR and PIR values specified. The interpretation of the administrative PIR is discussed below in Adaptation Rule.

Adaptation Rule

The adaptation rule provides the QoS provisioning system with the ability to adapt specific CIR and PIR defined administrative rates to the underlying capabilities of the hardware the queue will be created on to derive the operational rates. The administrative CIR and PIR rates are translated to actual operational rates enforced by the hardware queue. The rule provides a constraint used when the exact rate is not available due to hardware implementation trade-offs.

For the CIR and PIR parameters individually, the system will attempt to find the best operational rate depending on the defined constraint. The supported constraints are:

Depending on the hardware upon which the queue is provisioned, the actual operational CIR and PIR settings used by the queue will be dependent on the method the hardware uses to implement and represent the mechanisms that enforce the CIR and PIR rates.

As the hardware has a very granular set of rates, the recommended method to determine which hardware the rate is used for a given queue is to configure the queue rates with the associated adaptation rule and use the show pools output to display the actual rate achieved.

To illustrate how the adaptation rule constraints minimum, maximum and closest are evaluated in determining the operational CIR or PIR, assume there is a queue on an IOM3-XP where the administrative CIR and PIR values are 401 Mbps and 403 Mbps, respectively.

The following output shows the operating CIR and PIR rates achieved for the different adaptation rule settings:

Committed Burst Size

The committed burst size (CBS) parameters specify the amount of buffers that can be drawn from the reserved buffer portion of the queue’s buffer pool. Once the reserved buffers for a given queue have been used, the queue contends with other queues for additional buffer resources up to the maximum burst size.

The CBS is provisioned on ingress and egress service queues within service ingress QoS policies and service egress QoS policies, respectively. The CBS for a queue is specified in Kbytes.

The CBS for network queues are defined within network queue policies based on the forwarding class. The CBS for the queues for the forwarding class are defined as a percentage of buffer space for the pool.

Maximum Burst Size

The maximum burst size (MBS) parameter specifies the maximum queue depth to which a queue can grow. This parameter ensures that a customer that is massively or continuously over-subscribing the PIR of a queue will not consume all the available buffer resources. For high-priority forwarding class service queues, the MBS can be relatively smaller than the other forwarding class queues because the high-priority service packets are scheduled with priority over other service forwarding classes.

The MBS is provisioned on ingress and egress service queues within service ingress QoS policies and service egress QoS policies, respectively. The MBS for a queue is specified in Kbytes.

The MBS for network queues are defined within network queue policies based on the forwarding class. The MBS for the queues for the forwarding class are defined as a percentage of buffer space for the pool.

High Priority Only Buffers

High priority (HP) only buffers are defined on a queue and allow buffers to be reserved for traffic classified as high priority. When the queue depth reaches a specified level, only high-priority traffic can be enqueued. The HP only reservation for a queue is defined as a percentage of the MBS value.

On service ingress, the HP only reservation for a queue is defined in the service ingress QoS policy. High priority traffic is specified in the match criteria for the policy.

On service egress, the HP only reservation for a queue is defined in the service egress QoS policy. Service egress queues are specified by forwarding class. High priority traffic for a given traffic class is traffic that has been marked as in-profile either on ingress classification or based on interpretation of the QoS markings.

The HP only for network queues are defined within network queue policies based on the forwarding class. High priority traffic for a specific traffic class is marked as in-profile either on ingress classification or based on interpretation of the QoS markings.

Hi-Low Priority Only Buffers

Hi-low priority only buffers are defined on a queue and allow buffers to be reserved for traffic classified as high or low priority. When the queue depth reaches a specified level, only high and low priority traffic can be enqueued. The high/low priority only reservation for a queue is defined as a percentage of the MBS value. This is available for service egress and egress queue group queues. The high/low priority traffic for a given traffic class is traffic that has been determined to be in-profile or out-of-profile.

Hi-low priority only buffers exist also for egress network queues in a network queue policy, however, in these policies it is not configurable and has a default of an additional 10% of the MBS value on top of the high priority only.

WRED Per Queue

Egress SAP, subscriber, and network queues by default use drop tails within the queues and WRED slopes applied to the pools in which the queues reside in order to apply congestion control to the traffic in those queues. An alternative to this is to apply the WRED slopes directly to the egress queue using WRED per queue.

WRED per queue is supported for SAP egress QoS policy queues (and therefore to egress SAP and subscriber queues) and for queues within an egress queue group template. There are two modes available:

Native Queue Mode

When an egress queue is configured for native mode, it will use the native WRED capabilities of the forwarding plane queue. This is only supported on FP3 hardware.

Congestion control within the queue will use the low and exceed slopes from the applied slope policy together with the MBS drop tail. The queue continues to take buffers from its associated egress access or network buffer pool, on which WRED can also be enabled. This is shown in Figure 1.

Figure 1:  WRED queue: native mode 

To configure a native WRED queue, the wred-queue command is used under queue in a SAP egress QoS policy or egress queue group template with the mode set to native as follows.

Example:

qos

 queue-group-templates

  egress

   queue-group <queue-group-name> create

    queue <queue-id> create

     wred-queue [policy <slope-policy-name>] mode native

      slope-usage exceed-low

    exit

   exit

  exit

 exit

 sap-egress <policy-id> create

  queue <queue-id> [<queue-type>] [create]

   wred-queue [policy <slope-policy-name>] mode native

    slope-usage exceed-low

  exit

 exit

Congestion control is provided by both the slope policy applied to the queue and the MBS tail drop.

The slope-usage defines the mapping of the traffic profile to the WRED slope and only exceed-low is allowed with a native mode queue. The slope mapping is shown below and requires the low and exceed slopes to be no shutdown in the applied slope policy (otherwise traffic will use the MBS drop tail or a pool slope):

1. Out of profile maps to the low slope.
2. Exceed profile maps to the exceed slope.

The instantaneous queue depth is used against the slopes when native mode is configured, consequently the time-average-factor within the slope policy is ignored.

The in profile traffic uses the MBS drop tail for congestion control (the high slope is not used with a native mode queue).

If a queue is configured to use native mode WRED per queue on hardware earlier than FP3, the queue operates as a regular queue.

If the following SAP egress QoS policy is applied to SAP on an FP3 with the egress:

Example:

sap-egress 10 create

    queue 1 create

        wred-queue policy "slope1" mode native

                   slope-usage exceed-low

    exit

exit

the details of both the pool and queue can then be shown using this command:

*A:PE# show  pools access-egress 2/1/2 service 1 

===============================================================================

Pool Information

===============================================================================

Port                 : 2/1/2            

Application          : Acc-Egr          Pool Name             : default

CLI Config. Resv CBS : 30%(default)     

Resv CBS Step        : 0%               Resv CBS Max          : 0%

Amber Alarm Threshold: 0%               Red Alarm Threshold   : 0%

-------------------------------------------------------------------------------

Queue-Groups

-------------------------------------------------------------------------------

Queue-Group:Instance : policer-output-queues:1

-------------------------------------------------------------------------------

Utilization                   State       Start-Avg    Max-Avg    Max-Prob

-------------------------------------------------------------------------------

High-Slope                    Down              70%        90%         80%

Low-Slope                     Down              50%        75%         80%

Exceed-Slope                  Down              30%        55%         80%

Time Avg Factor      : 7                

Pool Total           : 74880 KB         

Pool Shared          : 51840 KB         Pool Resv             : 23040 KB

-------------------------------------------------------------------------------

-------------------------------------------------------------------------------

Current Resv CBS    Provisioned    Rising         Falling        Alarm

%age                all Queues     Alarm Thd      Alarm Thd      Color

-------------------------------------------------------------------------------

30%                 0 KB           NA             NA             Green

Pool Total In Use    : 0 KB             

Pool Shared In Use   : 0 KB             Pool Resv In Use      : 0 KB

WA Shared In Use     : 0 KB             

 

Hi-Slope Drop Prob   : 0                Lo-Slope Drop Prob    : 0

Excd-Slope Drop Prob : 0                

===============================================================================

Queue : 1->2/1/2:1->1

===============================================================================

FC Map           : be l2 af l1 h2 ef h1 nc

Tap              : not-applicable       

Admin PIR        : 10000000             Oper PIR         : Max

Admin CIR        : 0                    Oper CIR         : 0

Admin MBS        : 12240 KB             Oper MBS         : 12240 KB

Hi Prio Only     : not-applicable       Hi Low Prio Only : not-applicable

CBS              : 0 KB                 Depth            : 0

Slope            : slope1

WRED Mode        : native               Slope Usage      : exceed-low

-------------------------------------------------------------------------------

High Slope Information

-------------------------------------------------------------------------------

State            : disabled             

Start Average    : 0 KB                 Max Average      : 0 KB

Max Probability  : 0 %                  Curr Probability : 0 %

-------------------------------------------------------------------------------

-------------------------------------------------------------------------------

Low Slope Information

-------------------------------------------------------------------------------

State            : enabled              

Start Average    : 7200 KB              Max Average      : 9120 KB

Max Probability  : 80 %                 Curr Probability : 0 %

-------------------------------------------------------------------------------

                                      

-------------------------------------------------------------------------------

Exceed Slope Information

-------------------------------------------------------------------------------

State            : enabled              

Start Average    : 4680 KB              Max Average      : 6600 KB

Max Probability  : 80 %                 Curr Probability : 0 %

-------------------------------------------------------------------------------

===============================================================================

===============================================================================

*A:PE#

Pool per Queue Mode

When pool per queue mode is used, the queue resides in its own pool which is located in the forwarding plane egress megapool. The size of the pool is the same as the size of the queue (based on the MBS), consequently the WRED slopes operating on the pool's buffer utilization are reacting to the congestion depth of the queue. The size of the reserved CBS portion of the buffer pool is dictated by the queue's CBS parameter. This is shown in Figure 2.

Figure 2:  WRED queue: pool-per-queue mode 

The queue pools take buffers from the WRED egress megapool which must be enabled per FP; if this megapool is not enabled, the queue operates as a regular queue. By default only the Ingress normal and Egress normal megapools exist on an FP. The Egress WRED megapool is configured using the following commands:

Example:

configure

 card <slot-number>

  fp <fp-number>

   egress

    wred-queue-control

     [no] buffer-allocation min <percentage> max <percentage>

     [no] resv-cbs min <percentage> max <percentage>

     [no] shutdown

     [no] slope-policy <slope-policy-name>

The buffer allocation determines how much of the Egress normal megapool is allocated for the Egress WRED megapool, with the resv-cbs defining the amount of reserved buffers in the Egress WRED megapool. In both cases the min and max values must be equal. The slope-policy defines the WRED slope parameters and the time average factor used on the megapool itself to handle congestion as the megapool buffers are used. The no shutdown command enables the megapool. The megapools on card 1 FP 1 can be shown as follows:

*A:PE# show megapools 1 fp 1

===============================================================================

MegaPool Summary

===============================================================================

Type    Id        App.    Pool Name                   Actual ResvCBS  PoolSize

                                                      Admin ResvCBS

-------------------------------------------------------------------------------

fp      1/1       Egress  normal                      n/a             282624

                                                      n/a

fp      1/1       Ingress normal                      n/a             374784

                                                      n/a

fp      1/1       Egress  wred                        24576           93696

                                                      25%

===============================================================================

*A:PE#

To configure a pool per queue, the wred-queue command is used under queue in a SAP egress QoS policy or egress queue group template with the mode set to pool-per-queue as follows:

Example:

qos

 queue-group-templates

  egress

   queue-group <queue-group-name> create

    queue <queue-id> create

     wred-queue [policy <slope-policy-name>] mode pool-per-queue

       slope-usage default

    exit

   exit

  exit

 exit

 sap-egress <policy-id> create

  queue <queue-id> [<queue-type>] [create]

   wred-queue [policy <policy-name>] mode pool-per-queue

    slope-usage default

  exit

 exit

Congestion control is provided by the slope policy applied to the queue, with the slopes to be used having been no shutdown (otherwise traffic will use the MBS drop tail or a megapool slope). The slope-usage defines the mapping of the traffic profile to the WRED slope and only default is allowed with pool-per-queue which gives the following mapping:

1. In profile maps to the high slope.
2. Out of profile maps to the low slope.
3. Exceed profile maps to the exceed slope.

If the following SAP egress QoS policy is applied to SAP on an FP with the egress WRED megapool enabled:

Example:

sap-egress 10 create

    queue 1 create

       wred-queue policy "slope1" mode pool-per-queue

           slope-usage default

    exit

exit

the details of both the megapool and queue pool usages can then be shown using this command (detailed version):

*A:PE# show megapools 2 fp 1 wred service-id 1 detail 

===============================================================================

MegaPool Information

===============================================================================

Slot                 : 2                 

Application          : Egress            Pool Name          : wred

Resv CBS             : 25%               

-------------------------------------------------------------------------------

Utilization                   State       Start-Avg    Max-Avg    Max-Prob

-------------------------------------------------------------------------------

High-Slope                    Down              70%        90%         80%

Low-Slope                     Down              50%        75%         80%

Exceed-Slope                  Down              30%        55%         80%

Time Avg Factor      : 7                 

Pool Total           : 476160 KB         

Pool Shared          : 353280 KB         Pool Resv          : 122880 KB

Pool Total In Use    : 0 KB              

Pool Shared In Use   : 0 KB              Pool Resv In Use   : 0 KB

WA Shared In Use     : 0 KB              

 

Hi-Slope Drop Prob   : 0                 Lo-Slope Drop Prob : 0

Excd-Slope Drop Prob : 0                 

===============================================================================

Queue : 1->2/1/2:1->1

===============================================================================

FC Map           : be l2 af l1 h2 ef h1 nc

Admin PIR        : 10000000             Oper PIR         : Max

Admin CIR        : 0                    Oper CIR         : 0

Admin MBS        : 12240 KB             Oper MBS         : 12240 KB

Hi Prio Only     : not-applicable       Hi Low Prio Only : not-applicable

CBS              : 0 KB                 Depth            : 0

Slope            : slope1

WRED Mode        : pool-per-queue       Slope Usage      : default

Time Avg Factor  : 7                    

-------------------------------------------------------------------------------

High Slope Information

-------------------------------------------------------------------------------

State            : enabled              

Start Average    : 8880 KB              Max Average      : 10800 KB

Max Probability  : 80 %                 Curr Probability : 0 %

-------------------------------------------------------------------------------

                                      

-------------------------------------------------------------------------------

Low Slope Information

-------------------------------------------------------------------------------

State            : enabled              

Start Average    : 7200 KB              Max Average      : 9120 KB

Max Probability  : 80 %                 Curr Probability : 0 %

-------------------------------------------------------------------------------

-------------------------------------------------------------------------------

Exceed Slope Information

-------------------------------------------------------------------------------

State            : enabled              

Start Average    : 4800 KB              Max Average      : 6720 KB

Max Probability  : 80 %                 Curr Probability : 0 %

-------------------------------------------------------------------------------

-------------------------------------------------------------------------------

Pool Information

-------------------------------------------------------------------------------

Pool Total Allowed   : 12240 KB         

Pool Shared Allowed  : 12240 KB         Pool Resv             : 0 KB

High Slope Start Avg : 8880 KB          High slope Max Avg    : 10800 KB

Low Slope Start Avg  : 7200 KB          Low slope Max Avg     : 9120 KB

Excd Slope Start Avg : 4800 KB          Excd slope Max Avg    : 6720 KB

Pool Total In Use    : 0 KB             

From MegaPool Shared : 0 KB             From MegaPool Resv    : 0 KB

WA Pool In Use       : 0 KB             

 

Hi-Slope Drop Prob   : 0                Lo-Slope Drop Prob    : 0

Excd-Slope Drop Prob : 0                

-------------------------------------------------------------------------------

===============================================================================

===============================================================================

*A:PE#

Each WRED pool-per-queue uses a WRED pool resource on the FP. The resource usage can be seen in the tools dump resource-usage card [slot-num] fp [fp-number] output under Dynamic Q2 WRED Pools.

Packet Markings

Typically, customer markings placed on packets are not treated as trusted from an in-profile or out-of-profile perspective. This allows the use of the ingress buffering to absorb bursts over PIR from a customer and only perform marking as packets are scheduled out of the queue (as opposed to using a hard policing function that operates on the received rate from the customer). The resulting profile (in or out) based on ingress scheduling into the switch fabric is used by network egress for tunnel marking and egress congestion management.

The high/low priority feature allows a provider to offer a customer the ability to have some packets treated with a higher priority when buffered to the ingress queue. If the queue is configured with a hi-prio-only setting (setting the high priority MBS threshold higher than the queue’s low priority MBS threshold) a portion of the ingress queue’s allowed buffers are reserved for high priority traffic. An access ingress packet must hit an ingress QoS action in order for the ingress forwarding plane to treat the packet as high priority (the default is low priority).

If the packet’s ingress queue is above the low priority MBS, the packet will be discarded unless it has been classified as high priority. The priority of the packet is not retained after the packet is placed into the ingress queue. Once the packet is scheduled out of the ingress queue, the packet will be considered in-profile or out-of-profile based on the dynamic rate of the queue relative to the queue’s CIR parameter.

If an ingress queue is not configured with a hi-prio-only parameter, the low priority and high priority MBS thresholds will be the same. There will be no difference in high priority and low priority packet handling. At access ingress, the priority of a packet has no affect on which packets are scheduled first. Only the first buffering decision is affected. At ingress and egress, the current dynamic rate of the queue relative to the queue’s CIR does affect the scheduling priority between queues going to the same destination (either the switch fabric tap or egress port). The strict operating priority for queues are (from highest to lowest):

For access ingress, the CIR controls both dynamic scheduling priority and marking threshold. At network ingress, the queue’s CIR affects the scheduling priority but does not provide a profile marking function (as the network ingress policy trusts the received marking of the packet based on the network QoS policy).

At egress, the profile of a packet is only important for egress queue buffering decisions and egress marking decisions, not for scheduling priority. The egress queue’s CIR determines the dynamic scheduling priority, but does not affect the packet’s ingress determined profile.

Queue Counters

The router maintains counters for queues within the system for granular billing and accounting. Each queue maintains the following counters:

Queue-Types

The expedite, best-effort and auto-expedite queue types are mutually exclusive to each other. Each defines the method that the system uses to service the queue from a hardware perspective. While parental virtual schedulers can be defined for the queue, they only enforce how the queue interacts for bandwidth with other queues associated with the same scheduler hierarchy. An internal mechanism that provides access rules when the queue is vying for bandwidth with queues in other virtual schedulers is also needed.

Color Aware Profiling

The normal handling of SAP ingress access packets applies an in-profile or out-of-profile state to each packet relative to the dynamic rate of the queue as the packet is forwarded towards the egress side of the system. When the queue rate is within or equal to the configured CIR, the packet is considered in-profile. When the queue rate is above the CIR, the packet is considered out-of-profile. (This applies when the packet is scheduled out of the queue, not when the packet is buffered into the queue.) Egress queues use the profile marking of packets to preferentially buffer in-profile packets during congestion events. Once a packet has been marked in-profile or out-of-profile by the ingress access SLA enforcement, the packet is tagged with an in-profile or out-of-profile marking allowing congestion management in subsequent hops towards the packet’s ultimate destination. Each hop to the destination must have an ingress table that determines the in-profile or out-of-profile nature of a packet based on its QoS markings.

Color aware profiling adds the ability to selectively treat packets received on a SAP as in-profile or out-of-profile regardless of the queue forwarding rate. This allows a customer or access device to color a packet out-of-profile with the intention of preserving in-profile bandwidth for higher priority packets. The customer or access device may also color the packet in-profile, but this is rarely done as the original packets are usually already marked with the in-profile marking.

Each ingress access forwarding class may have one or multiple sub-class associations for SAP ingress classification purposes. Each sub-class retains the chassis wide behavior defined to the parent class while providing expanded ingress QoS classification actions. Sub-classes are created to provide a match association that enforces actions different than the parent forwarding class. These actions include explicit ingress remarking decisions and color aware functions.

All non-profiled and profiled packets are forwarded through the same ingress access queue to prevent out-of-sequence forwarding. Profiled packets in-profile are counted against the total packets flowing through the queue that are marked in-profile. This reduces the amount of CIR available to non-profiled packets causing fewer to be marked in-profile. Profiled packets out-of-profile are counted against the total packets flowing through the queue that are marked in-profile. This ensures that the amount of non-profiled packets marked out-of-profile is not affected by the profiled out-of-profile packet rate.

Service versus Network QoS

The QoS mechanisms within the routers are specialized for the type of traffic on the interface. For customer interfaces, there is service ingress and egress traffic, and for network core interfaces, there is network ingress and network egress traffic, as shown in Figure 3.

Figure 3:  Service vs Network Traffic Types 

The router uses QoS policies applied to a SAP for a service or to a network MDA/port to define the queuing, queue attributes, and QoS marking/interpretation.

The router supports four types of service and network QoS policies:

QoS Policy Entities

Services are configured with default QoS policies. Additional policies must be explicitly created and associated. There is one default service ingress QoS policy, one default service egress QoS policy, and one default network QoS policy. Only one ingress QoS policy and one egress QoS policy can be applied to a SAP or network port.

When you create a new QoS policy, default values are provided for most parameters with the exception of the policy ID and queue ID values, descriptions, and the default action queue assignment. Each policy has a scope, default action, a description, and at least one queue. The queue is associated with a forwarding class.

Service QoS policies can be applied to the following service types:

Network QoS policies can be applied to the following entities:

Network QoS policies can be applied to:

Default QoS policies maps all traffic with equal priority and allow an equal chance of transmission (Best Effort (be) forwarding class) and an equal chance of being dropped during periods of congestion. QoS prioritizes traffic according to the forwarding class and uses congestion management to control access ingress, access egress, and network traffic with queuing according to priority

Network QoS Policies

Network QoS policies define egress QoS marking and ingress QoS interpretation for traffic on core network IP interfaces. The router automatically creates egress queues for each of the forwarding classes on network IP interfaces.

A network QoS policy defines both the ingress and egress handling of QoS on the IP interface. The following functions are defined.

The required elements to be defined in a network QoS policy are:

Optional network QoS policy elements include:

Network policy ID 1 is reserved as the default network QoS policy. The default policy cannot be deleted or changed.

The default network QoS policy is applied to all network interfaces which do not have another network QoS policy explicitly assigned.

For network ingress, Table 6 and Table 7 list the default mapping of DSCP name and LSP EXP values to forwarding class and profile state for the default network QoS policy.

Network Queue QoS Policies

Network queue policies define the network forwarding class queue characteristics. Network queue policies are applied on egress on core network ports or channels and on ingress on MDAs. Network queue policies can be configured to use as many queues as needed. This means that the number of queues can vary. Not all policies will use eight queues like the default network queue policy.

The queue characteristics that can be configured on a per-forwarding class basis are:

Network queue policies are identified with a unique policy name which conforms to the standard router alphanumeric naming conventions.

The system default network queue policy is named default and cannot be edited or deleted. Table 7 describes the default network queue policy definition.

Service Ingress QoS Policies

Service ingress QoS policies define ingress service forwarding class queues and map flows to those queues. When a service ingress QoS policy is created by default, it always has two queues defined that cannot be deleted: one for the default unicast traffic and one for the default multipoint traffic. These queues exist within the definition of the policy. The queues only get instantiated in hardware when the policy is applied to a SAP. In the case where the service does not have multipoint traffic, the multipoint queues will not be instantiated.

In the simplest service ingress QoS policy, all traffic is treated as a single flow and mapped to a single queue, and all flooded traffic is treated with a single multipoint queue. The required elements to define a service ingress QoS policy are:

Optional service ingress QoS policy elements include:

To facilitate more forwarding classes, sub-classes are now supported. Each forwarding class can have one or multiple sub-class associations for SAP ingress classification purposes. Each sub-class retains the chassis wide behavior defined to the parent class while providing expanded ingress QoS classification actions.

There can now be up to 64 classes and subclasses combined in a sap-ingress policy. With the extra 56 values, the size of the forwarding class space is more than sufficient to handle the various combinations of actions.

Forwarding class expansion is accomplished through the explicit definition of sub-forwarding classes within the SAP ingress QoS policy. The CLI mechanism that creates forwarding class associations within the SAP ingress policy is also used to create sub-classes. A portion of the sub-class definition directly ties the sub-class to a parent, chassis wide forwarding class. The sub-class is only used as a SAP ingress QoS classification tool, the sub-class association is lost once ingress QoS processing is finished.

Figure 4:  Traffic Queuing Model for 3 Queues and 3 Classes 

When configured with this option, the forwarding class and drop priority of incoming traffic will be determined by the mapping result of the EXP bits in the top label. Table 8 displays he new classification hierarchy based on rule type.:

FC Mapping Based on EXP Bits at VLL/VPLS SAP

Figure 5:  Example Configuration — Carrier’s Carrier Application 

To accommodate backbone ISPs who want to provide VPLS/VLL to small ISPs as a site-to-site inter-connection service, small ISP routers can connect to Ethernet Layer 2 SAPs. The traffic will be encapsulated in a VLL/VPLS SDP. These small ISP routers are typically PE router. In order to provide appropriate QoS, the SR-Series router support a new classification option that based on received MPLS EXP bits.

The lsp-exp command is will be supported in sap-ingress qos policy. This option can only be applied on Ethernet Layer 2 SAPs.

Table 9:  Forwarding Class Classification Based on Rule Type   

#	Rule	Forwarding Class	Comments
1	default-fc	Set the policy’s default forwarding class.	All packets match the default rule.
2	IP criteria: Multiple entries per policy Multiple criteria per entry	Set when an fc-name exists in the entry’s action. Otherwise, preserve from the previous match.	When IP criteria is specified, entries are matched based on ascending order until first match and then processing stops. A packet can only match a single IP criteria entry.
3	MAC criteria: Multiple entries per policy Multiple criteria per entry	Set when an fc-name exists in the entry’s action. Otherwise, preserve from the previous match.	When MAC criteria is specified, entries are matched based on ascending order until first match and then processing stops. A packet can only match a single MAC criteria entry.

The enqueuing priority is specified as part of the classification rule and is set to “high” or “low”. The enqueuing priority relates to the forwarding class queue’s High-Priority-Only allocation where only packets with a high enqueuing priority are accepted into the queue once the queue’s depth reaches the defined threshold. (See High Priority Only Buffers.)

The mapping of IEEE 802.1p bits, IP Precedence and DSCP values to forwarding classes is optional as is specifying IP and MAC criteria.

The IP and MAC match criteria can be very basic or quite detailed. IP and MAC match criteria are constructed from policy entries. An entry is identified by a unique, numerical entry ID. A single entry cannot contain more than one match value for each match criteria. Each match entry has a queuing action which specifies: the forwarding class of packets that match the entry.

1. The forwarding class of packets that match the entry
2. The enqueuing priority (high or low) for matching packets

The entries are evaluated in numerical order based on the entry ID from the lowest to highest ID value. The first entry that matches all match criteria has its action performed.

The supported service ingress QoS policy IP match criteria are:

1. Destination IP address/prefix
2. Destination port/range
3. IP fragment
4. Protocol type (TCP, UDP, etc.)
5. Source port/range
6. Source IP address/prefix
7. DSCP value

The supported service ingress QoS policy MAC match criteria are:

1. IEEE 802.2 LLC SSAP value/mask
2. IEEE 802.2 LLC DSAP value/mask
3. IEEE 802.3 LLC SNAP OUI zero or non-zero value
4. IEEE 802.3 LLC SNAP PID value
5. IEEE 802.1p value/mask
6. Source MAC address/mask
7. Destination MAC address/mask
8. EtherType value

The MAC match criteria that can be used for an Ethernet frame depends on the frame’s format. See Table 10.

Table 10:  MAC Match Ethernet Frame Types   

Frame Format	Description
802dot3	IEEE 802.3 Ethernet frame. Only the source MAC, destination MAC and IEEE 802.1p value are compared for match criteria.
802dot2-llc	IEEE 802.3 Ethernet frame with an 802.2 LLC header.
802dot2-snap	IEEE 802.2 Ethernet frame with 802.2 SNAP header.
Ethernet-II	Ethernet type II frame where the 802.3 length field is used as an Ethernet type (Etype) value. Etype values are two byte values greater than 0x5FF (1535 decimal).

The 802dot3 frame format matches across all Ethernet frame formats where only the source MAC, destination MAC and IEEE 802.1p value are compared. The other Ethernet frame types match those field values in addition to fields specific to the frame format. Table 11 lists the criteria that can be matched for the various MAC frame types.

Table 11:  MAC Match Criteria Frame Type Dependencies   

Frame Format	Source MAC	Dest MAC	IEEE 802.1p Value	Etype Value	LLC Header SSAP/DSAP Value/Mask	SNAP-OUI Zero/Non-zero Value	SNAP- PID Value
802dot3	Yes	Yes	Yes	No	No	No	No
802dot2-llc	Yes	Yes	Yes	No	Yes	No	No
802dot2-snap	Yes	Yes	Yes	No	No 1	Yes	Yes
ethernet-II	Yes	Yes	Yes	Yes	No	No	No

Note:

1. When a SNAP header is present, the LLC header is always set to AA-AA

Service ingress QoS policy ID 1 is reserved for the default service ingress policy. The default policy cannot be deleted or changed.

The default service ingress policy is implicitly applied to all SAPs which do not explicitly have another service ingress policy assigned. The characteristics of the default policy are listed in Table 12.

Table 12:  Default Service Ingress Policy ID 1 Definition   

Characteristic	Item	Definition
Queues	Queue 1	1 (one) queue all unicast traffic: Forward Class: best-effort (be) CIR = 0 PIR = max (line rate) MBS, CBS and HP Only = default (values derived from applicable policy)
	Queue 11	1 (one) queue for all multipoint traffic: CIR = 0 PIR = max (line rate) MBS, CBS and HP Only = default (values derived from applicable policy)
Flows	Default Forwarding Class	1 (one) flow defined for all traffic: All traffic mapped to best-effort (be) with a low priority

Egress Forwarding Class Override

Egress forwarding class override provides additional QoS flexibility by allowing the use of a different forwarding class at egress than was used at ingress.

The ingress QoS processing classifies traffic into a forwarding class (or sub-class) and by default the same forwarding class is used for this traffic at the access or network egress. The ingress forwarding class or sub-class can be overridden so that the traffic uses a different forwarding class at the egress. This can be configured for the main forwarding classes and for sub-classes, allowing each to use a different forwarding class at the egress.

The buffering, queuing, policing and remarking operation at the ingress and egress remain unchanged. Egress reclassification is possible. The profile processing is completely unaffected by overriding the forwarding class.

When used in conjunction with QPPB (QoS Policy Propagation Using BGP), a QPPB assigned forwarding class takes precedence over both the normal ingress forwarding class classification rules and any egress forwarding class overrides.

Figure 6:  Egress Forwarding Class Override 

Figure 6 shows the ingress service 1 using forwarding classes AF and L1 that are overridden to L1 for the network egress, while it also shows ingress service 2 using forwarding classes L1, AF, and L2 that are overridden to AF for the network egress.

Service Egress QoS Policies

Service egress queues are implemented at the transition from the service core network to the service access network. The advantages of per-service queuing before transmission into the access network are:

The subrate capabilities and per-service scheduling control are required to make multiple services per physical port possible. Without egress shaping, it is impossible to support more than one service per port. There is no way to prevent service traffic from bursting to the available port bandwidth and starving other services.

For accounting purposes, per-service statistics can be logged. When statistics from service ingress queues are compared with service egress queues, the ability to conform to per-service QoS requirements within the service core can be measured. The service core statistics are a major asset to core provisioning tools.

Service egress QoS policies define egress queues and map forwarding class flows to queues. In the simplest service egress QoS policy, all forwarding classes are treated like a single flow and mapped to a single queue. To define a basic egress QoS policy, the following are required:

Optional service egress QoS policy elements include:

Each queue in a policy is associated with one of the forwarding classes. Each queue can have its individual queue parameters allowing individual rate shaping of the forwarding class(es) mapped to the queue.

More complex service queuing models are supported in the router where each forwarding class is associated with a dedicated queue.

The forwarding class determination per service egress packet is determined either at ingress or egress. If the packet ingressed the service on the same router, the service ingress classification rules determine the forwarding class of the packet. If the packet is received on a network interface, the forwarding class is marked in the tunnel transport encapsulation. In each case, the packet can be reclassified into a different forwarding class at service egress.

Service egress QoS policy ID 1 is reserved as the default service egress policy. The default policy cannot be deleted or changed. The default access egress policy is applied to all SAPs service egress policy explicitly assigned. The characteristics of the default policy are listed in the following table.

Named Pool Policies

The named buffer pool feature for the 7450 ESS and 7750 SR allows for the creation of named buffer pools at the MDA and port level. Named pools allow for a customized buffer allocation mode for ingress and egress queues that goes beyond the default pool behavior.

Named pools are defined within a named pool policy. The policy contains a q1-pools context which is used to define port allocation weights and named pools for buffer pools on Q1 based IOMs (all IOMs that are currently supported). The policy may be applied at either the port or MDA level at which time the pools defined within the policy are created on the port or MDA. When the policy is applied at the MDA level, MDA named pools are created. MDA named pools will typically be used when either a pool cannot be created per port or when the buffering needs of queues mapped to the pool are not affected by sharing the pool with queues from other ports. MDA named pools allow buffers to be efficiently shared between queues on different ports mapped to the same pool. However, MDA named pools do present the possibility that very active queues on one port could deplete buffers in the pool offering the possibility that queues on other ports experiencing buffer starvation. Port named pools are created when the policy is applied at the port level and allow for a more surgical application of the buffer space allocated for a physical port. MDA pool names do not need to be unique. If a name overlaps exists, the port pool will be used. The same pool name may be created on multiple ports on the same MDA.

The named pool policy is applied at the MDA ingress and egress level and at the ingress and egress port level. Each MDA within the system is associated with a forwarding plane traffic manager that has support for a maximum of 57 buffer pools. The following circumstances affect the number of named pools that can be created per MDA (these circumstances may be different between ingress and egress for the MDA):

If the named pool policy is applied to an MDA or port that cannot create every pool defined in the policy, the policy application attempt will fail. Any preexisting named pool policy on the MDA or port will not be affected by the failed named pool policy association attempt.

When buffer pools are being created or deleted, individual queues may need to be moved to or from the default pools. When a queue is being moved, the traffic destined to the queue is first moved temporarily to a ‘fail-over’ queue. Then the queue is allowed to drain. Once the queue is drained, the statistics for the queue are copied. The queue is then returned to the free queue list. A new queue is then created associated with the appropriate buffer pool, the saved stats are loaded to the queue and then the traffic is moved from the fail-over queue to the new queue. While the traffic is being moved between the old queue to the fail-over queue and then to the new queue, some out of order forwarding may be experienced. Also, any traffic forwarded through the fail-over queue will not be accounted for in billing or accounting statistics. A similar action is performed for queues that have the associated pool name added, changed or removed. This only applies to where fail-over queues are currently supported.

The first step in allowing named pools to be created for an MDA is to enable ‘named-pool-mode’ at the IOM level (config card slot-number named-pool-mode). Named pool mode may be enabled and disabled at anytime. When MDAs are currently provisioned on the IOM, the IOM is reset to allow all existing pools to be deleted and the new default, named MDA and named port pools to be created and sized. If MDAs are not currently provisioned (as when the system is booting up), the IOM is not reset. When named pool mode is enabled, the system changes the way that default pools are created. The system no longer creates default pools per port, instead, a set of per forwarding plane level pools are created that are used by all queues that are not explicitly mapped to a named pool.

After the IOM has been placed into named pool mode, a named pool policy must be associated with the ingress and egress contexts of the MDA or individual ports on the MDA for named pools to be created. There are no named pools that exist by default.

Each time the default pool reserve, aggregate MDA pool limit or individual pool sizes is changed, buffer pool allocation must be re-evaluated.

Pools may be deleted from the named pool policy at anytime. Queues associated with removed or non-existent pools are mapped to one of the default pools based on whether the queue is access or ingress. The queue is flagged as ‘pool-orphaned’ until either the pool comes into existence, or the pool name association is changed on the pool.

An ingress or egress port managed buffer space is derived from the port’s active bandwidth. Based on this bandwidth value compared to the other port’s bandwidth value, the available buffer space is given to each port to manage. It may be desirable to artificially increase or decrease this bandwidth value to compensate for how many buffers are actually needed on each port. If one port has very few queues associated with it and another has many queues associated, the commands in the port’s “modify-buffer-allocation-rate” CLI context may be used to move one port’s bandwidth up, and another port’s bandwidth down. As provisioning levels change between ports, the rate modification commands may be used to adapt the buffer allocations per port.

Buffer allocation rate modification is supported for both standard and named pool mode buffer allocation methods.

The system allocates buffers based on the following criteria:

Slope Policies

For network ingress, a buffer pool is created for the XMA or MDA and is used for all network ingress queues for ports on the XMA or MDA.

Slope policies define the RED slope characteristics as a percentage of pool size for the pool on which the policy is applied.

Default buffer pools exist (logically) at the port and XMA and MDA levels. Each physical port has two pool objects associated:

By default, each pool is associated with slope-policy default.

Access and network pools (in network mode) and access uplink pools (in access uplink mode) are created at the port level; creation is dependent on the physical port mode (network, access) or the mode of provisioned channel paths.

Node-level pools are used by ingress network queues and bundle access queues. A single ingress network pool is created at the node-level for ingress network queues.

An ingress and egress access pool is created at the MDA level for all bundle access queues.

Slope policies can also be applied when using WRED per Queue, see WRED Per Queue.

Tuning the Shared Buffer Utilization Calculation

The router allows tuning the calculation of the Shared Buffer Average Utilization (SBAU) after assigning buffers for a packet entering a queue as used by the RED slopes to calculate a packet’s drop probability. The router implements a time average factor (TAF) parameter in the buffer policy which determines the contribution of the historical shared buffer utilization and the instantaneous Shared Buffer Utilization (SBU) in calculating the SBAU. The TAF defines a weighting exponent used to determine the portion of the shared buffer instantaneous utilization and the previous shared buffer average utilization used to calculate the new shared buffer average utilization. To derive the new shared buffer average utilization, the buffer pool takes a portion of the previous shared buffer average and adds it to the inverse portion of the instantaneous shared buffer utilization (SBU). The formula used to calculated the average shared buffer utilization is:

Figure 8:  Equation 

where:

SBAUn = Shared buffer average utilization for event n

SBAUn-1 = Shared buffer average utilization for event (n-1)

SBU = The instantaneous shared buffer utilization

TAF = The time average factor

Table 14 shows the effect the allowed values of TAF have on the relative weighting of the instantaneous SBU and the previous SBAU (SBAUn-1) has on the calculating the current SBAU (SBAUn).

Table 14:  TAF Impact on Shared Buffer Average Utilization Calculation   

TAF	2TAF	Equates To	Shared Buffer Instantaneous Utilization Portion	Shared Buffer Average Utilization Portion
0	20	1	1/1 (1)	0 (0)
1	21	2	1/2 (0.5)	1/2 (0.5)
2	22	4	1/4 (0.25)	3/4 (0.75)
3	23	8	1/8 (0.125)	7/8 (0.875)
4	24	16	1/16 (0.0625)	15/16 (0.9375)
5	25	32	1/32 (0.03125)	31/32 (0.96875)
6	26	64	1/64 (0.015625)	63/64 (0.984375)
7	27	128	1/128 (0.0078125)	127/128 (0.9921875)
8	28	256	1/256 (0.00390625)	255/256 (0.99609375)
9	29	512	1/512 (0.001953125)	511/512 (0.998046875)
10	210	1024	1/1024 (0.0009765625)	1023/2024 (0.9990234375)
11	211	2048	1/2048 (0.00048828125)	2047/2048 (0.99951171875)
12	212	4096	1/4096 (0.000244140625)	4095/4096 (0.999755859375)
13	213	8192	1/8192 (0.0001220703125)	8191/8192 (0.9998779296875)
14	214	16384	1/16384 (0.00006103515625)	16383/16384 (0.99993896484375)
15	215	32768	1/32768 (0.000030517578125)	32767/32768 (0.999969482421875)

The value specified for the TAF affects the speed at which the shared buffer average utilization tracks the instantaneous shared buffer utilization. A low value weights the new shared buffer average utilization calculation more to the shared buffer instantaneous utilization. When TAF is zero, the shared buffer average utilization is equal to the instantaneous shared buffer utilization. A high value weights the new shared buffer average utilization calculation more to the previous shared buffer average utilization value. The TAF value applies to all high and low priority RED slopes for ingress and egress buffer pools controlled by the buffer policy.

Scheduler Policies

A scheduler policy defines the hierarchy and all operating parameters for the member schedulers. A scheduler policy must be defined in the QoS context before a group of virtual schedulers can be used. Although configured in a scheduler policy, the individual schedulers are actually created when the policy is applied to a site, such as a SAP or interface.

Scheduler objects define bandwidth controls that limit each child (other schedulers and queues) associated with the scheduler. The scheduler object can also define a child association with a parent scheduler of its own.

A scheduler is used to define a bandwidth aggregation point within the hierarchy of virtual schedulers. The scheduler’s rate defines the maximum bandwidth that the scheduler can consume. It is assumed that each scheduler created will have queues or other schedulers defined as child associations. The scheduler can also be a child (take bandwidth from) a scheduler in a higher tier, except for schedulers created in Tier 1.

A parent parameter can be defined to specify a scheduler further up in the scheduler policy hierarchy. Only schedulers in Tiers 2 and 3 can have parental association. Tier 1 schedulers cannot have a parental association. When multiple schedulers and/or queues share a child status with the scheduler on the parent, the weight or strict parameters define how this scheduler contends with the other children for the parent’s bandwidth. The parent scheduler can be removed or changed at anytime and is immediately reflected on the schedulers actually created by association of this scheduler policy.

When a parent scheduler is defined without specifying level, weight, or CIR parameters, the default bandwidth access method is weight with a value of 1.

If any orphaned queues (queues specifying a scheduler name that does not exist) exist on the ingress SAP and the policy application creates the required scheduler, the status on the queue becomes non-orphaned at this time.

Figure 9 depicts how child queues and schedulers interact with their parent scheduler to receive bandwidth. The scheduler distributes bandwidth to the children by first using each child’s CIR according to the CIR-level parameter (CIR L8 through CIR L1 weighted loops). The weighting at each CIR-Level loop is defined by the CIR weight parameter for each child. The scheduler then distributes any remaining bandwidth to the children up to each child’s rate parameter according to the Level parameter (L8 through L1 weighted loops). The weighting at each level loop is defined by the weight parameter for each child.

Figure 9:  Virtual Scheduler Internal Bandwidth Allocation 

Tiers

In single tier scheduling, queues are scheduled based on the forwarding class of the queue and the operational state of the queue relative to the queue’s CIR and PIR. Queues operating within their CIR values are serviced before queues operating above their CIR values with “high-priority” forwarding class queues given preference over “low-priority” forwarding class queues. In single tier scheduling, all queues are treated as if they are at the same “level” and the queue’s parameters and operational state directly dictate the queue’s scheduling. Single tier scheduling is the system default scheduling policy for all the queues and destinations listed above and has no configurable parameters.

Hierarchical scheduler policies are an alternate way to schedule queues that can be used on service ingress and service egress queues. Hierarchical scheduler policies allow the creation of a hierarchy of schedulers where queues and/or other schedulers are scheduled by superior schedulers.

To illustrate the difference between single tier scheduling and hierarchical scheduling policies, consider a simple case where, on service ingress, three queues are created for gold, silver and bronze service and are configured as follows:

1. Gold: CIR = 10 Mbps, PIR = 10 Mbps
2. Silver: CIR = 20 Mbps, PIR = 40 Mbps
3. Bronze: CIR = 0 Mbps, PIR = 100 Mbps

In the router, the CIR is used for profiling of traffic (in-profile or out-of-profile), and the PIR is the rate at which traffic is shaped out of the queue. In single tier scheduling, each queue can burst up to its defined PIR, which means up to 150 Mbps (10 Mbps + 40 Mbps + 100 Mbps) can enter the service.

In a simple example of a hierarchical scheduling policy, a superior (or parent) scheduler can be created for the gold, silver and bronze queues which limits the overall rate for all queues to 100 Mbps. In this hierarchical scheduling policy, the customer can send in any combination of gold, silver and bronze traffic conforming to the defined PIR values and not to exceed 100 Mbps.

Single Tier Scheduling

Single-tier scheduling is the default method of scheduling queues in the router. Queues are scheduled with single-tier scheduling if no explicit hierarchical scheduler policy is defined or applied. There are no configurable parameters for single-tier scheduling.

In single tier scheduling, queues are scheduled based on the Forwarding Class of the queue and the operational state of the queue relative to the queue’s Committed Information Rate (CIR) and Peak Information Rate (PIR). Queue’s operating within their CIR values are serviced before queue’s operating above their CIR values with “high-priority” forwarding class queues given preference over “low-priority” forwarding class queues. In Single Tier Scheduling, all queues are treated as if they are at the same “level” and the queue’s parameters and operational state directly dictate the queue’s scheduling.

A pair of schedulers, a high-priority and low-priority scheduler, transmits to a single destination switch fabric port, access port, or network interface. Table 17 below lists how the forwarding class queues are mapped to the high and low scheduler:

Table 17:  Forwarding Class Scheduler Mapping  

Scheduler	Forwarding Class
High	Network Control
	Expedited
	High-2
	High 1
Low	Low-1
	Assured
	Low-2
	Best-Effort

By using the default QoS profile, all ingress traffic is treated as best effort (be) (mapped to FC be and to low priority scheduler). For an egress SAP using the default QoS profile, all egress traffic will use the same queue.

While competing for bandwidth to the destination, each scheduler determines which queue will be serviced next. During congestion (packets existing on multiple queues), queues are serviced in the following order:

Queues associated with the high-priority scheduler operating within their CIR

1. Queues associated with the low-priority scheduler operating within their CIR
2. All queues with traffic above CIR and within PIR will be serviced by a biased round robin

Queues associated with a single scheduler are serviced in a round robin method. If a queue reaches the configured PIR, the scheduler will not serve the queue until the transmission rate drops below the PIR.

The router QoS features are flexible and allow modifications to the forwarding class characteristics and the CIR and PIR queue parameters. The only fundamental QoS mechanisms enforced within the hardware are the association of the forwarding classes with the high priority or low priority scheduler and the scheduling algorithm. Other parameters can be modified to configure the appropriate QoS behavior.

Hierarchical Scheduler Policies

Hierarchical scheduler policies are an alternate way of scheduling queues which can be used on service ingress and service egress queues. Hierarchical scheduler policies allow the creation of a hierarchy of schedulers where queues and/or other schedulers are scheduled by superior schedulers.

The use of the hierarchical scheduler policies is often referred to as hierarchical QoS or H-QoS on the SR OS.

Hierarchical Virtual Schedulers

Virtual schedulers are created within the context of a hierarchical scheduler policy. A hierarchical scheduler policy defines the hierarchy and parameters for each scheduler. A scheduler is defined in the context of a tier (Tier 1, Tier 2, Tier 3). The tier level determines the scheduler’s position within the hierarchy. Three tiers of virtual schedulers are supported (Figure 10). Tier 1 schedulers (also called root schedulers) are defined without a parent scheduler. It is not necessary for Tier 1 schedulers to obtain bandwidth from a higher tier scheduler. A scheduler can enforce a maximum rate of operation for all child queues and associated schedulers.

Figure 10:  Hierarchical Scheduler and Queue Association 

Scheduler Policies Applied to Applications

A scheduler policy can be applied either on a SAP (Figure 11) or on a multi-service customer site (a group of SAPs with common origination/termination point) (Figure 12). Whenever a scheduler policy is applied, the individual schedulers comprising the policy are created on the object. When the object is an individual SAP, only queues created on that SAP can use the schedulers created by the policy association. When the object is a multi-service customer site, the schedulers are available to any SAPs associated with the site (also see Scheduler Policies Applied to SAPs).

Figure 11:  Scheduler Policy on SAP and Scheduler Hierarchy Creation 

Figure 12:  Scheduler Policy on Customer Site and Scheduler Hierarchy Creation 

Queues become associated with schedulers when the parent scheduler name is defined within the queue definition in the SAP ingress policy. The scheduler is used to provide bandwidth to the queue relative to the operating constraints imposed by the scheduler hierarchy.

Shared Queues

Shared-queue QoS policies can be implemented to reduce ingress queue consumption on an MDA. It is especially useful when VPLS, IES, and VPRN services are scaled on one MDA. Instead of allocating multiple hardware queues for each unicast or multipoint queue defined in a SAP ingress QoS policy, SAPs with the shared-queuing feature enabled only allocate one hardware queue for each SAP ingress QoS policy unicast or multipoint queue.

However, as a tradeoff, the total amount of traffic throughput at ingress of the node is reduced because any ingress packet serviced by a shared-queuing SAP is recirculated for further processing. When the node is only used for access SAPs, 5 Gbps ingress traffic is the maximum that can be processed without seeing packet drops at the MDA ingress. The reason for this is that any ingress packet serviced by a shared-queuing SAP is processed twice in the forwarding plane which greatly reduces bandwidth.

Shared-queuing can add latency. Network planners should consider these restrictions while trying to scale services on one MDA.

ATM Traffic Descriptor Profiles

Traffic descriptors profiles capture the cell arrival pattern for resource allocation. Source traffic descriptors for an ATM connection include at least one of the following:

QoS Traffic descriptor profiles are applied on IES, VPRN, VPLS, and VLL SAPs.

ATM traffic descriptors are not supported on the 7950 XRS.

Configuration Notes

The following information describes QoS implementation caveats: