Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Configuring Large Delay Buffers in CoS

You can configure very large delay buffers using the buffer-size-temporal command combined with the q-pic-large-buffer command. The buffer-size temporal option in combination with q-pic-large-buffer can create extra-large delay buffer allocations for one or several queues on an interface.

Note:

If the configured buffer size is too low, the buffer size for the forwarding class defaults to 9192 and the following log message is displayed: “fwdd_cos_set_delay_bandwidth:queue:16 delay buffer size (1414) too low, setting to default 9192.”

Configuring Large Delay Buffers

The following configuration applies to the examples that follow:

  1. Configure two VLANs (one ingress, one egress) on one interface. No interface shaping rate is initially defined for this configuration.
  2. Enable the q-pic-large-buffer option on the same PIC, in addition to the buffer-size temporal option on the queue, to create a large buffer on the queue:
    Note:

    The CLI does not provide a warning when you use buffer-size temporal without q-pic-large-buffer. When you use buffer-size temporal, verify that the configuration also includes the q-pic-large buffer command.

  3. Define four forwarding-classes (queue names) for the four queues:
  4. Configure the forwarding classes (queue names) included in a scheduler map, applied to the egress VLAN:
  5. Set the queue priorities. Only queue priorities are initially defined, not transmit rates or buffer sizes.

Example: Simple Configuration Using Four Queues

This configuration allocates 12,500,000 bytes of buffer for each of the four queues. To avoid exceeding the limits of the delay buffer calculation, this initial example has no interface shaping rate, scheduler transmit rate, or scheduler buffer size percent configuration.

  1. Specify the maximum 4-second delay buffer on each of the four queues:

    Specifying buffer-size temporal on some or all queues uses implicit (default) or explicit transmit rate percentages as the buffer-size percentages of the temporal values for those queues. Because there are no explicitly specified transmit rate percentages, divide 100 percent by the number of configured queues (queues with schedulers configured in the scheduler map) to get the implicit (default) per-queue transmit rate percentages. Each queue gets an implicit (default) transmit rate of 100% / 4 = 25%.

    In this example, specifying the maximum 4-second delay on each queue, with no shaping rate on the interface and implicit (default) per-queue transmit rates of 25 percent, the total buffer for all temporal 4m queues on an interface = 4 seconds * 100,000,000 maximum interface bps / 8 bits/byte = 4 seconds * 12,500,000 bytes = 50,000,000 bytes. Each queue specifying temporal 4m gets 25% * 50,000,000 = 12,500,000 bytes.

  2. Add a shaping rate of 4 Mbps to the interface:

    The total buffer for all temporal 4m queues on an interface = 4 sec * 4,000,000 bps shaping-rate / 8 bits/byte = 4 sec * 500,000 bytes = 2,000,000 bytes. Therefore, each queue specifying temporal 4m receives 25% * 2,000,000 = 500,000 bytes.

When using buffer-size temporal on any interface queues, if you also use the transmit-rate percent command, or the buffer-size percent command, or both commands, on any of the interface queues, the buffer size calculations become more complex and the limits of available queue depth might be reached. If the configuration attempts to exceed the available memory, then at commit time two system log messages appear in the /var/log/messages file, the interface class-of-service configuration is ignored, and the interface class-of-service configuration reverts to the two-queue defaults:

When configuring buffer-size temporal along with transmit-rate percent or buffer-size percent, or both, you must monitor the system log to see whether the available queue depth limit has been reached.

Example: Using buffer-size temporal with Explicit transmit-rate percent Commands

To add explicit transmit rates to all four queues:

For example, if an interface is shaped to 4 Mbps, the transmit rate percentage of 10 for a queue means that the bandwidth share for the specific queue is 0.4 Mbps. The queues are allocated portions of the 2,000,000 bytes of total buffer available for temporal queues on this interface, proportionally to their transmit rates. The four queues get 200,000, 500,000, 500,000, and 800,000 bytes of delay buffer, respectively.

To avoid exceeding the queue depth limits and triggering system log messages and default configuration behavior, when configuring queues with buffer-size temporal and transmit rate percent and other (non-temporal) queues with buffer-size percent, the following configuration rule must be followed: When one or more queues on an interface are configured with buffer-size temporal, the sum of the temporal queues explicitly configured transmit rate percentages plus other non-temporal queues explicitly configured buffer size percentages must not exceed 100 percent.

If the total of the temporal queues transmit rate percentages and the non-temporal queues buffer-size percentages exceeds 100 percent, the queue mem underflow and Failed to compute scheduler params system log messages appear in the messages log, the explicitly configured CLI CoS configuration for the interface is ignored, and the interface reverts to a two-queue default CoS configuration.

When buffer-size temporal is specified on a queue, if transmit-rate percent is also configured on the same queue, the queue depth configured is based on the fractional bandwidth for the queue as obtained by the specified transmit-rate percent.

In addition to temporal delay times specified for one or more queues using buffer size temporal, there is another delay time automatically computed for the entire interface. This interface delay time is distributed across all non-temporal queues, proportionally to their implicit (default) or explicit transmit-rate percentages. If q-pic-large-buffer is not enabled, the interface delay time defaults to 100 ms. As shown in Table 1, when q-pic-large-buffer is enabled, interface delay time is calculated according to configured shaping rate for the interface. Because the shaping-rate configured in the example above was 4 Mbps (> 2,048,000 bps), the interface delay time for the configuration is 100 msec.

Table 1: Interface Delay Times Enabled By q-pic-large-buffer

Configured Shaping Rate (bps)

Interface Delay Time (msec) Used for Non-Temporal Queues with q-pic-large-buffer Enabled

Default Delay Time Used (msec) Without q-pic-large-buffer

64,000-255,999

4000

100

256,000 - 511,999

2000

100

512,000 - 102,3999

1000

100

1,024,000 - 2,047,999

500

100

>= 2,048,000

100

100

This example properly computes the delay buffer limits on both temporal and non-temporal queues:

  1. Substitute buffer-size percent for buffer-size temporal on queues 0 and 1:

    This deletes the requirement for hard-specified 4 seconds of buffering and replaces it with a proportional limit of 10 percent (or 25 percent) of the total interface delay time for the non-temporal queues. In both cases, the queue depth is calculated based on the share of the interface bandwidth for the specific queues. Total Interface Non-Temporal Queue Memory = shaping-rate * Interface delay time (Table 1) = 4 Mbps * 0.1 seconds = 500,000 bytes per second * 0.1 seconds = 50,000 bytes, therefore queues 0 and 1 get 10% * 50,000 = 5000 bytes and 25% * 50,000 = 12,500 bytes of delay buffer, respectively.

  2. Configure buffer-size temporal on queues 2 and 3:

    Queues 2 and 3 still get 500,000 and 800,000 bytes of delay buffer, respectively, as previously calculated. This configuration obeys the rule that the sum of the temporal queues transmit rate percentages (25% + 40% = 65%), plus the non-temporal queues buffer size percentages (10% + 25% = 35%) do not exceed 100% (65% + 35% <= 100%).

The following example exceeds the delay buffer limit, triggering the system log messages and the default, two-queue class-of-service behavior.

Increase the buffer-size percentage from 25 percent to 26 percent for non-temporal queue 1:

This violates the configuration rule that the sum of the non-temporal queues buffer-size percentages (10% + 26% = 36%), plus the temporal queues transmit rate percentages (25% + 40% = 65%) now exceed 100% (36% + 65% = 101%). Therefore, the following two system log messages appear in the /var/log/messages file:

When the delay buffer limits are exceeded, the CLI-configured class-of-service settings are not used and the default class-of-service configuration (the default scheduler-map) is assigned to the interface. This uses two queues: the forwarding-class best-effort (queue 0) has transmit rate percent 95 and buffer-size percent 95 and the forwarding-class network-control (queue 3) has the transmit rate percent 5 and buffer-size percent 5.

Related Documentation