Expand All close

Traffic Management User Guide for OCX Series Devices

play_arrow CoS Overview
- play_arrow Basic Concepts
- play_arrow Configuration Statements for Basic Concepts
  - class-of-service
  - traceoptions (Class of Service)
- play_arrow Monitoring Commands for Basic Concepts
play_arrow Classifying and Rewriting Traffic
play_arrow Scheduling Traffic
play_arrow Ethernet PAUSE
- play_arrow Using Ethernet PAUSE
  - Understanding CoS Flow Control (Ethernet PAUSE and PFC)
  - Enabling and Disabling CoS Symmetric Ethernet PAUSE Flow Control
- play_arrow Configuration Statements for Ethernet PAUSE
  - flow-control
play_arrow Buffers
- play_arrow Using Buffers
- play_arrow Configuration Statements for Buffers
- play_arrow Monitoring Commands for Buffers
  - show class-of-service shared-buffer
play_arrow Downloads
- Download this guide: Traffic Management User Guide for OCX Series Devices

list Table of Contents

PDF

English

Overview of CoS on OCX Series Switches

date_range 31-Aug-17

arrow_backward

arrow_forward

Many Juniper Networks data center switching platforms are optimized for Layer 2 Ethernet transport. However, OCX Series switches are optimized for Layer 3 IP transport. This difference results in some differences in class-of-service (CoS) feature support on OCX Series switches compared with, for example, QFX Series switches. This topic describes those differences.

In addition, some Juniper Networks documentation is shared among different Juniper Networks platforms. Because of this, you might see references to lossless transport, data center bridging exchange (DCBX) protocol, priority-based flow control (PFC), and Fibre Channel over Ethernet (FCoE) in the documentation. These references do not apply to OCX Series switches.

Supported CoS Features

OCX Series switches support the following CoS features:

Incoming packet classification on Layer 3 physical interfaces when at least one logical interface is defined on the physical interface:
- DSCP, DSCP IPv6, and IEEE 802.1p behavior aggregate (BA) classifiers
- Fixed classifiers
- Multifield classifiers
The default DSCP BA classifier is the default classifier. It maps incoming unicast traffic into the best-effort (queue 0) and network-control (queue 7) forwarding classes.
Up to eight unicast forwarding classes and up to four multidestination (multicast, broadcast, destination lookup fail) forwarding classes.
DSCP, DSCP IPv6, and IEEE 802.1p rewrite rules on Layer 3 physical interfaces when at least one logical interface is defined on the physical interface.
Hierarchical, two-tier port scheduling, also known as enhanced transmission selection (ETS).
Per output queue control of forwarding classes:
- Guaranteed minimum bandwidth
- Maximum bandwidth
- Scheduling priority
- Weighted random early detection (WRED) packet drop characteristics for congestion management
Explicit congestion notification (ECN).
Symmetric Ethernet PAUSE flow control.
Shared and dedicated buffer pool configuration.

Unsupported Lossless Transport Features

OCX Series switches do not support lossless transport. Lossless transport does not refer to best-effort traffic on a link with Ethernet PAUSE enabled. Lossless transport refers to traffic classified into lossless forwarding classes on which you enable priority-based flow control (PFC) (defined in IEEE 802.1Qbb). OCX Series switches do not support lossless forwarding classes and do not support PFC.

Unsupported Default Lossless Unicast Forwarding Classes

Because the Junos OS software is common to several data center switching platforms, two of the five default forwarding classes are lossless forwarding classes (the unicast fcoe and no-loss forwarding classes, which are mapped by default to output queue 3 and output queue 4, respectively). On OCX Series switches, the default fcoe and no-loss forwarding classes are not supported.

The default fcoe and no-loss lossless forwarding classes carry the no-loss packet drop attribute. On OCX Series switches, the no-loss packet drop attribute is not supported. Do not classify traffic into the default fcoe or no-loss forwarding classes. Do not configure a forwarding class with the no-loss packet drop attribute.

You can use queues 3 and 4, but you must configure a forwarding class and map it to the desired queue, configure a classifier to map incoming traffic to the forwarding class, and then apply the classifier to the appropriate interfaces.

The forwarding class names fcoe and no-loss are just that—names. If you want to carry traffic on queue 3 or queue 4, you can remove the no-loss packet drop attribute from these forwarding classes without changing the names. Or, you can change the forwarding class names to something else, just as long as you do not configure the no-loss packet drop attribute on the forwarding class.

For example, to configure a new forwarding class named be2, map it to queue 3, map traffic identified by DSCP code point 001010 to forwarding class be2, and apply the configuration to interface xe-0/0/20:

Configure a new forwarding class named be2, without the no-loss attribute, and map it to queue 3:
[edit class-of-service] user@switch# set forwarding-classes class be2 queue-num 3
Configure a classifier named be_classifier to classify incoming traffic with DSCP code point 001010 into forwarding class be2:
[edit class-of-service] user@switch# set classifiers dscp be_classifier import default forwarding-class be2 loss-priority low code-points 001010
Importing the default classifier bases the new classifier on the default classifier. The default packet classification is retained, except for the changes you make to the classifier configuration. In this case, you are adding a new mapping to the classifier default mapping and saving it as classifier be_classifier.
Apply the classifier to interface xe-0/0/20:
[edit class-of-service] user@switch# set interfaces xe-0/0/20 classifiers dscp be_classifier

Note

In addition to configuring the forwarding class and packet classification, you must also add the new forwarding class to a forwarding class set, and apply the forwarding class set to interfaces in order to associate the traffic in a forwarding class with interfaces. For completeness, here are example statements to accomplish this

This brief addition to the example configures a forwarding class set named be_fc_set, maps forwarding class be2 and default forwarding class best-effort to be_fc_set, and applies be_fc_set to interface xe-0/0/20:

[edit class-of-service]
user@switch# set forwarding-class-sets be_fc_set class be2
user@switch# set forwarding-class-sets be_fc_set class
best-effort
user@switch# set interfaces xe-0/0/20 forwarding-class-set be_fc_set

Because this example classifies traffic into a forwarding class that is mapped to one of the queues that the default scheduler services, traffic receives the default scheduling (bandwidth and priority) for that queue. If you classify traffic into a forwarding class that is mapped to a queue that does not receive default scheduling, configure a queue scheduler for the traffic to ensure that the traffic receives a minimum amount of bandwidth during periods of congestion.

Effect of Unsupported Unicast Lossless Forwarding Classes on Bandwidth Scheduling

The default scheduler provides bandwidth and scheduling priority for unicast queues as shown in Table 1:

Table 1: Default Scheduler Configuration

Default Scheduler and Queue Number	Transmit Rate (Minimum Guaranteed Bandwidth)	Priority
best-effort forwarding class scheduler (queue 0)	5%	low
fcoe forwarding class scheduler (queue 3)	35%	low
no-loss forwarding class scheduler (queue 4)	35%	low
network-control forwarding class scheduler (queue 7)	5%	low
mcast forwarding class scheduler (queue 8)	20%	low

Because the default DSCP classifier maps unicast traffic only into queue 0 and queue 7, the 35 percent of the bandwidth allocated to queue 3 and queue 4 are shared with the traffic-bearing queues. However, if you create a new forwarding class and map it to queue 3 or queue 4, the traffic classified into that forwarding class receives the scheduling resources shown in Table 1. If you want to change the default scheduler bandwidth allocations, configure a hierarchical port scheduler.

If you want to map unicast traffic to queues other than the default queues, configure a hierarchical port scheduler to allocate port bandwidth resources to the queues.

Priority-Based Flow Control

OCX Series switches do not support PFC, and they do not support configuration of the [edit class-of-service congestion-notification-profile] hierarchy in the CLI or applying a congestion notification profile to an interface. In the rest of the CoS documentation set, information about PFC does not apply to OCX Series switches.

DCBX

OCX Series switches do not support DCBX, and they do not support configuration of the [edit class-of-service congestion-notification-profile] hierarchy in the CLI. In the rest of the CoS documentation set, information about DCBX does not apply to OCX Series switches.

Unsupported CoS Features

OCX Series switches do not support the following CoS features that some other data center switches support:

CoS on Layer 2 interfaces (OCX Series switches do not support family ethernet-switching interfaces).
Incoming MPLS EXP packet classification (OCX Series switches do not support MPLS and do not support the [edit class-of-service system-defaults classifiers] hierarchy).
Lossless transport.
Default lossless forwarding classes (fcoe and no-loss forwarding classes).
No-loss packet drop attribute; do not apply the no-loss packet drop attribute to any forwarding class, and do not use the default fcoe and no-loss forwarding classes, which carry the no-loss packet drop attribute by default.
Priority-based flow control (PFC).
Data center bridging exchange (DCBX) protocol.
Outgoing MPLS EXP rewrite rules (OCX Series switches do not support MPLS).
Asymmetric Ethernet PAUSE flow control.