[Contents] [Prev] [Next] [Index] [Report an Error]

Overview

QoS is a suite of features that configure queuing and scheduling on the forwarding path of the E-series router. QoS provides a level of predictability and control beyond the best-effort delivery that the router provides by default. Best-effort service provides packet transmission with no assurance of reliability, delay, jitter, or throughput.

QoS as developed for E-series routers conforms to the IETF Differentiated Services (DiffServ) model (RFCs 2597 and 2598). DiffServ networks classify packets into one of a small number of aggregated flows or traffic classes for which you can configure different QoS characteristics. The Juniper Networks QoS architecture extends DiffServ to support edge features such as high-density queuing.

The E-series router supports:

IETF architecture for differentiated services
Assured forwarding per-hop-behavior (PHB) groups
Expedited forwarding PHB groups

See References for a list of related RFCs.

The router supports configurable queuing and scheduling. It has an application-specific integrated circuit (ASIC) scheduler that supports thousands of queues in a hierarchical round-robin (HRR) scheduler. The scheduler allows the router to allocate separate queues for each forwarding interface. Separate queues enable fair access to buffers and bandwidth for each subscriber connected to the router.

Allocating queues per interface allows an Internet service provider (ISP) to shape an individual subscriber's traffic flows to specified rates independent of the underlying Layer 2 network type.

The E-series router supports QoS on the 5-, 10-, and 40-Gbps fabric boards. It supports egress line module functions only on ASIC-based line modules.

Figure 3 shows the traffic flow through the router.

Figure 3: Traffic Flow Through an E-series Router

Terms

Table 16 defines terms used in this discussion of QoS.




 Table 16:  QoS Terminology Used in This Chapter 

 Term
 Description

 Assured rate

 Bandwidth guaranteed until oversubscribed.


 Best effort

 Network forwards as many packets as possible in as reasonable a time as possible. This is the default per-hop behavior (PHB) for packet transmission.


 Best-effort queue

 For a logical interface, the queue associated with the best-effort traffic class for that logical interface,


 Best-effort scheduler node

 The scheduler node associated with a logical interface and traffic class group pair, and where the traffic class group contains the best-effort traffic class. Also known as best-effort node.


 CDV

 Cell delay variation. Measures the difference between a cell's expected and actual transfer delay. Determines the amount of jitter.


 CDVT

 Cell delay variation tolerance. Specifies the acceptable tolerance of CDV (jitter).


 Effective weight

 The result of a weight or an assured rate. Users configure the scheduler node by specifying either an assured rate or a weight within a scheduler profile. An assured rate, in bits per second, is translated into a weight. The resultant weight is referred to as an effective weight. 


 Group node

 A scheduler node associated with a {port interface, traffic-class group} pair. Because the logical interface is the port, only one such scheduler node can exist for each traffic-class group above the port. This node aggregates all traffic for traffic classes in the group.


 HAR

 Hierarchical assured rate. Dynamically adjusts bandwidth for scheduler nodes.


 HRR

 Hierarchical round-robin. Allocates bandwidth to queues in proportion to their weights.


 Latency

 Delay in the transmission of a packet through a network from beginning to end.


 Proprietary QoS Management Information Base (MIB)

 Supported on the E-series router.


 Queue

 First-in-first-out (FIFO) set of buffers that control packets on the data path.


 QoS port-type profile

 Supplies the QoS information for forwarding interfaces stacked above ports of the associated interface type.


 QoS profile attachment

 Applies the rules in the QoS profile to a specific interface. 


 Rate shaping

 Allows you to throttle a queue to a specified rate.


 RED

 Random early detection congestion avoidance technique.


 Scheduler hierarchy

 A hierarchical, tree-like arrangement of scheduler nodes and queues. The router supports up to three levels of scheduler nodes stacked above a port (level 0), with a final level of queues stacked above the nodes. A traffic-class group uses a scheduler level at level 1.


 Scheduler node

 An element within the hierarchical scheduler that implements bandwidth controls for a group of queues. Queues are stacked above scheduler nodes in a hierarchy. The root node is associated with a channel or physical port.


 Shared shaper constituent

 All nodes and queues that are associated with a logical interface that is being shared shaped are considered potential constituents of the shared shaper. 


 Weight

 Specifies the relative weight for queues in the traffic class. 


 WRED

 Weighted random early detection congestion avoidance technique.

Term	Description
Assured rate	Bandwidth guaranteed until oversubscribed.
Best effort	Network forwards as many packets as possible in as reasonable a time as possible. This is the default per-hop behavior (PHB) for packet transmission.
Best-effort queue	For a logical interface, the queue associated with the best-effort traffic class for that logical interface,
Best-effort scheduler node	The scheduler node associated with a logical interface and traffic class group pair, and where the traffic class group contains the best-effort traffic class. Also known as best-effort node.
CDV	Cell delay variation. Measures the difference between a cell's expected and actual transfer delay. Determines the amount of jitter.
CDVT	Cell delay variation tolerance. Specifies the acceptable tolerance of CDV (jitter).
Effective weight	The result of a weight or an assured rate. Users configure the scheduler node by specifying either an assured rate or a weight within a scheduler profile. An assured rate, in bits per second, is translated into a weight. The resultant weight is referred to as an effective weight.
Group node	A scheduler node associated with a {port interface, traffic-class group} pair. Because the logical interface is the port, only one such scheduler node can exist for each traffic-class group above the port. This node aggregates all traffic for traffic classes in the group.
HAR	Hierarchical assured rate. Dynamically adjusts bandwidth for scheduler nodes.
HRR	Hierarchical round-robin. Allocates bandwidth to queues in proportion to their weights.
Latency	Delay in the transmission of a packet through a network from beginning to end.
Proprietary QoS Management Information Base (MIB)	Supported on the E-series router.
Queue	First-in-first-out (FIFO) set of buffers that control packets on the data path.
QoS port-type profile	Supplies the QoS information for forwarding interfaces stacked above ports of the associated interface type.
QoS profile attachment	Applies the rules in the QoS profile to a specific interface.
Rate shaping	Allows you to throttle a queue to a specified rate.
RED	Random early detection congestion avoidance technique.
Scheduler hierarchy	A hierarchical, tree-like arrangement of scheduler nodes and queues. The router supports up to three levels of scheduler nodes stacked above a port (level 0), with a final level of queues stacked above the nodes. A traffic-class group uses a scheduler level at level 1.
Scheduler node	An element within the hierarchical scheduler that implements bandwidth controls for a group of queues. Queues are stacked above scheduler nodes in a hierarchy. The root node is associated with a channel or physical port.
Shared shaper constituent	All nodes and queues that are associated with a logical interface that is being shared shaped are considered potential constituents of the shared shaper.
Weight	Specifies the relative weight for queues in the traffic class.
WRED	Weighted random early detection congestion avoidance technique.

Features

Table 17 describes the major QoS features that the E-series router provides.




 Table 17:  QoS Features 

 Feature
 Description

 Best effort

 Default traffic class for packets being forwarded across the device. Packets that are not assigned to a specific traffic class are assigned to the best-effort traffic class.


 Differentiated services


   Assured forwarding—See RFC 2597.
   Expedited forwarding—See RFC 2598.



 Drop profile

 Template that specifies active queue management in the form of WRED behavior of an egress queue. 


 Port shaping

 Shapes the aggregate traffic through a port or channel to a rate that is less than the line or port rate. 


 QoS port-type profile

 QoS profile that is automatically attached to ports of the corresponding type if you do not explicitly attach a QoS profile.


 QoS profile

 Collection of QoS commands that specify queue profiles, drop profiles, scheduler profiles, and statistics profiles in combination with interface types.


 Queue profile

 Template that specifies the buffering and tail-dropping behavior of an egress queue.


 Rate shaping

 Mechanism that throttles the rate at which an interface can transmit packets.
 Note: Rate shaping as presented in policy management in releases before JUNOSe 4.0 is deprecated and converted to QoS profiles and scheduler profiles. 


 Relative strict-priority scheduling

 Provides strict-priority scheduling within a shaped aggregate rate. For example, it lets you provide 1 Mbps of aggregate bandwidth to a subscriber, with up to 500 Kbps of the bandwidth for low-latency traffic. If there is no strict-priority traffic, the low-latency traffic can use up to the full aggregate rate of 1 Mbps.


 Scheduler profile

 Configures the bandwidth at which queues drain as a function of relative weight, assured rate, and shaping rate.


 Shared rate shaping

 Mechanism that enables dynamic sharing of logical interface bandwidth for traffic that is queued through separate scheduler hierarchies.


 Statistics profile

 Template that specifies rate statistics and event-gathering characteristics. 


 Strict-priority scheduling

 Designates the traffic class (queue) that receives top priority for transmission of its packets through a port. It is implemented with a special strict-priority scheduler node that is stacked directly above the port. 


 Traffic class

 A chassis-wide grouping of queues and buffers that support transmission of a designated set of traffic across the chassis, from ingress line module, through the switch fabric, and onto the egress line module.
 The router supports up to eight traffic classes, and therefore up to eight queues per logical interface.


 Traffic-class group

 Separate hierarchy of scheduler nodes and queues over a port. A traffic-class group uses one level of the scheduler hierarchy, level 1.
 Traffic classes belong to the default group unless they are specifically assigned to a named group. All queues are stacked in a single scheduler hierarchy above the physical port. When you configure a traffic class inside a group, its queues are stacked separately. The most common reason for creating separate scheduler hierarchies is to implement strict priority scheduling for all queues in the group.
 The router supports up to four traffic-class groups. A traffic class cannot belong to more than one group.


 WRED

 Signals end-to-end protocols such as TCP that the router is becoming congested along a particular egress path. The intent is to trigger TCP congestion avoidance in a random set of TCP flows before congestion becomes severe and causes tail dropping on a large number of flows.

Feature	Description
Best effort	Default traffic class for packets being forwarded across the device. Packets that are not assigned to a specific traffic class are assigned to the best-effort traffic class.
Differentiated services	Assured forwarding—See RFC 2597. Expedited forwarding—See RFC 2598.
Drop profile	Template that specifies active queue management in the form of WRED behavior of an egress queue.
Port shaping	Shapes the aggregate traffic through a port or channel to a rate that is less than the line or port rate.
QoS port-type profile	QoS profile that is automatically attached to ports of the corresponding type if you do not explicitly attach a QoS profile.
QoS profile	Collection of QoS commands that specify queue profiles, drop profiles, scheduler profiles, and statistics profiles in combination with interface types.
Queue profile	Template that specifies the buffering and tail-dropping behavior of an egress queue.
Rate shaping	Mechanism that throttles the rate at which an interface can transmit packets. Note: Rate shaping as presented in policy management in releases before JUNOSe 4.0 is deprecated and converted to QoS profiles and scheduler profiles.
Relative strict-priority scheduling	Provides strict-priority scheduling within a shaped aggregate rate. For example, it lets you provide 1 Mbps of aggregate bandwidth to a subscriber, with up to 500 Kbps of the bandwidth for low-latency traffic. If there is no strict-priority traffic, the low-latency traffic can use up to the full aggregate rate of 1 Mbps.
Scheduler profile	Configures the bandwidth at which queues drain as a function of relative weight, assured rate, and shaping rate.
Shared rate shaping	Mechanism that enables dynamic sharing of logical interface bandwidth for traffic that is queued through separate scheduler hierarchies.
Statistics profile	Template that specifies rate statistics and event-gathering characteristics.
Strict-priority scheduling	Designates the traffic class (queue) that receives top priority for transmission of its packets through a port. It is implemented with a special strict-priority scheduler node that is stacked directly above the port.
Traffic class	A chassis-wide grouping of queues and buffers that support transmission of a designated set of traffic across the chassis, from ingress line module, through the switch fabric, and onto the egress line module. The router supports up to eight traffic classes, and therefore up to eight queues per logical interface.
Traffic-class group	Separate hierarchy of scheduler nodes and queues over a port. A traffic-class group uses one level of the scheduler hierarchy, level 1. Traffic classes belong to the default group unless they are specifically assigned to a named group. All queues are stacked in a single scheduler hierarchy above the physical port. When you configure a traffic class inside a group, its queues are stacked separately. The most common reason for creating separate scheduler hierarchies is to implement strict priority scheduling for all queues in the group. The router supports up to four traffic-class groups. A traffic class cannot belong to more than one group.
WRED	Signals end-to-end protocols such as TCP that the router is becoming congested along a particular egress path. The intent is to trigger TCP congestion avoidance in a random set of TCP flows before congestion becomes severe and causes tail dropping on a large number of flows.

[Contents] [Prev] [Next] [Index] [Report an Error]