Rate and give feedback:  Feedback Received. Thank You!

Rate and give feedback: 

X

This document helped resolve my issue

Yes No

Additional Comments

800 characters remaining

May we contact you if necessary?

Name:  

E-mail: 

Enter a valid Email ID

Need product assistance? Contact Juniper Support

Submit

This site is protected by hCaptcha and its Privacy Policy and Terms of Service apply.

English  

English

Dedicated Queue Scaling for CoS Configurations on MIC and MPC Interfaces Overview

The 30-Gigabit Ethernet Queuing and 60-Gigabit Ethernet Queuing and Enhanced Queuing Ethernet Modular Port Concentrators (MPCs) provide a set of dedicated queues for subscriber interfaces configured with hierarchical scheduling or per-unit scheduling.

The dedicated queues offered on these MPCs enable service providers to reduce costs through different scaling configurations. For example, the 60-Gigabit Ethernet Enhanced Queuing MPC enables service providers to reduce the cost per subscriber by allowing many subscriber interfaces to be created with four or eight queues. Alternatively, the 30-Gigabit Ethernet and 60-Gigabit Ethernet Queuing MPCs enable service providers to reduce hardware costs, but allow fewer subscriber interfaces to be created with four or eight queues.

This topic describes the overall queue, scheduler node, and logical interface scaling for subscriber interfaces created on these MIC and MPC combinations.

Queue Scaling for MIC and MPC Combinations

Table 1 lists the number of dedicated queues and number of subscribers supported per MPC.

MPCs vary in the number of Packet Forwarding Engines on board. MPC1s, such as the 30-Gigabit Ethernet MPC, have one Packet Forwarding Engine. MPC2s, such as the 60-Gigabit Ethernet MPC, have two Packet Forwarding Engines. Each Packet Forwarding Engine has two schedulers that share the management of the queues.

A scheduler maps to one-half of a MIC; in CLI configuration statements, that one-half of a MIC corresponds to PIC 0, 1, 2, or 3. MIC ports are partitioned equally across the PICs. A two-port MIC has one port per PIC. A four-port MIC has two ports per PIC.

Each interface-set uses eight queues from total available egress queues.

Distribution of Queues on 30-Gigabit Ethernet Queuing MPCs

On 30-Gigabit Ethernet Queuing MPCs, each scheduler maps to different PICs. When only one MIC is installed, scheduler 0 maps to PIC 0 and scheduler 1 maps to PIC 1 on the MIC. When two MICs are installed, scheduler 0 can additionally distribute queues to PIC 2 on MIC 1, and scheduler 1 can additionally distribute queues to PIC 3 on MIC 1. However, the distribution of queues to the MICs is not hard-partitioned for 30-Gigabit Ethernet Queuing MPCs or other MPC1s. Distribution depends instead on how you allocate the queues to the PICs.

Figure 1 shows the queue distribution on a 30-Gigabit Ethernet Queuing MPC with only one MIC installed. All 64,000 egress queues on the MPC are available to the single Packet Forwarding Engine. On the Packet Forwarding Engine, half of these queues (32,000) are managed by each scheduler. Scheduler 0 contributes all of its 32,000 queues to PIC 0. Scheduler 1 contributes all of its 32,000 queues to PIC 1.

Figure 1: Distribution of Queues on the 30-Gigabit Ethernet Queuing MPC with One MIC

Figure 2 shows the queue distribution on the same MPC with two MICs installed. In this case, each scheduler can supply two PICS, one on each MIC. Because the distribution of the queues across the MICs is not hard-partitioned, you can allocate from 0 to 32,000 queues from each scheduler’s pool across the scheduler’s associated PICs. For example, you can allocate 32,000 queues from Scheduler 0 to PIC 0, 4000 queues from Scheduler 1 to PIC 1, and 28,000 queues from Scheduler 1 to PIC 3. Alternatively, you can allocate the queues evenly across the PICs, or allocate them in other combinations with the limitation of 32,000 queues per PIC and 32,000 queues per port.

Figure 2: Distribution of Queues on the 30-Gigabit Ethernet Queuing MPC with Two MICs

Distribution of Queues on 60-Gigabit Ethernet MPCs

On 60-Gigabit Ethernet Queuing and Enhanced Queuing Ethernet MPCs, each scheduler maps to a single PIC: PIC 0 or PIC 1 on MIC 0 and PIC 2 or PIC 3 on MIC 1. The distribution of the queues is hard-partitioned for these MPCs and other MPC2s; the only difference in distribution is in the total number of queues available.

For example, Figure 3 shows how queues are distributed on a 60-Gigabit Ethernet Enhanced Queuing MPC. Of the 512,000 egress queues on the MPC, half (256,000) are available to each of the two Packet Forwarding Engines. On each Packet Forwarding Engine, half of these queues (128,000) are managed by each scheduler. The complete scheduler complement (128,000) is available to only one PIC in a MIC. Thus the total number of queues available depends on the number of MICs installed. The MPC must have 2 MICs to achieve the maximum of 512,000 queues. With a single MIC, the MPC can achieve only 256,000 queues.

Figure 3: Distribution of Queues on the 60-Gigabit Ethernet Enhanced Queuing MPC

Determining Maximum Egress Queues and Subscriber Interfaces per Port

The number of MICs installed in an MPC and the number of ports per MIC do not affect the maximum number of queues available on a given port. These factors affect only how you are able to allocate queues (and, therefore, subscribers) for your network.

For example, a 30-Gigabit Ethernet Queuing MPC supports a maximum of 16,000 subscriber interfaces and has a maximum of 32,000 queues available per PIC. On this card, you can allocate up to 32,000 queues to a single port in each PIC. If you dedicate 4 queues per subscriber interface, you can accommodate a maximum of 8000 subscriber interfaces on a single port, and therefore need at least two ports to reach the maximum 16,000 subscriber interfaces. If you dedicate 8 queues per subscriber interface, you can accommodate a maximum of 4000 subscriber interfaces on a single port, and you need 4 ports for the maximum of 16,000 subscriber interfaces.

The 60-Gigabit Ethernet Enhanced Queuing MPC supports a maximum of 64,000 subscriber interfaces and has a maximum of 128,000 queues per PIC. You can allocate up to 128,000 queues to a single port in each PIC. However, if you dedicate 4 queues per subscriber interface, you can accommodate a maximum of only 16,000 subscriber interfaces on a single MPC port—not 32,000—because the 60-Gigabit Ethernet Enhanced Queuing MPC is limited to 16,000 subscriber interfaces per PIC. If you dedicate 8 queues per subscriber interface, you can also accommodate a maximum of 16,000 subscriber interfaces on a single MPC port. In either case, you need at least 4 ports to reach the maximum of 64,000 subscriber interfaces.

Managing Remaining Queues

When the number of available dedicated queues on the MPC drops below 10 percent, an SNMP trap is generated to notify you .

When the maximum number of dedicated queues on the MPCs is reached, a system log message, COSD_OUT_OF_DEDICATED_QUEUES, is generated. The system does not provide subsequent subscriber interfaces with a dedicated set of queues. For per-unit scheduling configurations, there are no configurable queues remaining on the MPC.

For hierarchical scheduling configurations, remaining queues are available when the maximum number of dedicated queues is reached on the MPC. Traffic from these logical interfaces are considered unclassified and attached to a common set of queues that are shared by all subsequent logical interfaces. These common queues are the default port queues that are created for every port. You can configure a traffic control profile and attach that to the interface to provide CoS parameters for the remaining queues.

For example, when the 30-Gigabit Ethernet Queuing MPC is configured with 32,000 subscriber interfaces with four queues per subscriber, the MPC can support 16,000 subscribers with a dedicated set of queues. You can provide CoS shaping and scheduling parameters to the remaining queues for those subscriber interfaces by attaching a special traffic-control profile to the interface.

These subscriber interfaces remain with this traffic control profile, even if dedicated queues become available.