- Cisco Community
- Technology and Support
- Networking
- Networking Knowledge Base
- Increased Total output drops/OutDiscards by QoS for Desktop Series Switches Products
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
- Subscribe to RSS Feed
- Mark as New
- Mark as Read
- Bookmark
- Subscribe
- Printer Friendly Page
- Report Inappropriate Content
04-06-2015 06:27 PM - edited 03-01-2019 05:05 PM
<Outline>
Every year, many TAC Service Requests are opened regarding problems of degradation of traffic forwarding performance as the Total output drops/OutDiscards counter of the Desktop Series Switches products is increased.
This document explains the problems of increase in the Total output drops/OutDiscards counter and degradation of traffic forwarding performance caused by configuring QoS for the Desktop Series Switches products.
Specifically, it applies to the following products:
Catalyst 2960 Series products
Catalyst 2960S Series products
Catalyst 2960X Series products
Catalyst 2960XR Series products
Catalyst 3750 Series products
Catalyst 3560 Series products
Catalyst 3750E Series products
Catalyst 3560E Series products
Catalyst 3750X Series products
Catalyst 3560X Series products
Total output drops/OutDiscards is counted when traffic is not stored in the transmit queue used for traffic forwarding, and discarded.
There are two primary causes of traffic being discarded without storing in the transmit queue as described below. As the first one is explained in another document, this document explains the latter one.
- Ingress traffic that exceeds a bandwidth of output interface
- QoS configuration
Reference)
Catalyst 3560, 3750 Series, Increase in OutDiscard Counter
https://supportforums.cisco.com/ja/document/12050011
<Increase in Total output drops/OutDiscards by QoS configuration>
QoS is a function to realize which traffic is given limited resources preferentially.
After configuring QoS, a state occurs where resources are preferentially allocated to certain traffic while other traffic does not have priority of resource allocation.
We are often asked whether Total output drops/OutDiscards is over-counted because only "mls qos" is configured and other QoSs are not configured. However, the resource allocation will never be optimized by configuring "mls qos" only.
When operating under the conditions described above, in most cases, Total output drops/OutDiscards occurs more often than in a case where "mls qos" is not configured, and the whole performance is degraded.
The following explains three major impacts that trigger the occurrence of Total output drops/OutDiscards caused by enabling QoS.
<Impact of increasing Queue>
Before and after configuring QoS with the "mls qos" command, the number of Queues used for user traffic forwarding differs as shown below:
The number of Queues before configuring QoS
The number of Queues for user traffic forwarding -> 1
The number of Queues after configuring QoS
The number of Queue for user traffic forwarding -> 4
In terms of user traffic forwarding, one Queue can monopolize the buffer before configuring QoS. After configuring QoS, however, because the buffer is shared by four Queues, if it is not shared properly, output drops/OutDiscards occur more frequently than before configuring QoS, and the traffic forwarding performance will be degraded.
Take a look at the following simplified example. When the following traffic is assigned to four Queues with default settings while QoS is enabled, most traffic is processed in Queue 2. With default settings, because a buffer assignment for Queue 2 is the same as a buffer assignment for other ports, output drops/OutDiscards are very likely to occur in this Queue while buffer is wasted in Queue 1 and Queue 4, as traffic is not assigned in these Queues.
Queue 1->
Queue 2->Traffic A, B, C, D
Queue 3->Traffic E
Queue 4->
For the reasons above, which traffic is assigned to each Queue needs to be decided carefully.
Traffic which is assigned to each Queue can be checked by the "show mls qos maps" command.
Traffic is assigned to Queue by the "mls qos srr-queue output" command.
<Impact of limiting buffer to use for Queue>
Before and after configuring QoS with the "mls qos" command, limitations on buffer that can be used for Queue differ, in addition to the difference in the number of Queues described above.
When QoS is not enabled, in terms of user traffic forwarding, one Queue can use all buffers for user forwarding traffic.
On the other hand, for the default setting with enabling QoS, a maximum amount of buffer that can be used by each Queue is set to a considerably low value compared with the state where QoS is disabled.
This default setting avoids problems occurring from other Queues by one Queue monopolizing many buffers, and it assumes that tuning is implemented in accordance with usage conditions.
After enabling QoS, if tuning in accordance with the usage conditions is not implemented, output drops/OutDiscards are very likely to occur, degrading the traffic forwarding performance.
Buffer conditions above in each Queue and thresholds are checked by the "show mls qos queue-set" command.
Adjustment is performed by the "mls qos queue-set output" command.
The settings below enable each Queue to use buffers as much as possible.
TAC often informs you of this setting as the output drops/OutDiscards measures; this is due to the reasons above.
mls qos queue-set output 1 threshold 1 3200 3200 100 3200
mls qos queue-set output 1 threshold 2 3200 3200 100 3200
mls qos queue-set output 1 threshold 3 3200 3200 100 3200
mls qos queue-set output 1 threshold 4 3200 3200 100 3200
<Impact of 1/25 Shaping of Queue 1>
Shaping by 1/25 is set to Queue 1 by enabling QoS.
This means almost the same, as the band width is limited by 1/25; it is limited to about 4M if it is a 100M link, and it is limited to about 40M if it is a 1G link.
Queue 1 gets into a state where it continuously stores a large volume of traffic due to decrease in speed to be transmitted as traffic is taken out from Queue 1, making Queue 1 unable to store new traffic. As a result, many output drops/OutDiscards occur in Queue 1, which degrades traffic forwarding performance for processing to be performed in this Queue.
Among cases in which output drops/OutDiscards increase after enabling QoS, many of them are caused by the reason described above.
The shaping in the above is checked by the "show mls qos interface queueing" command.
It is set by the "srr-queue bandwidth shape" command.
The below setting disables this 1/25 shaping:
srr-queue bandwidth share 0 0 0 0
TAC also often informs you of this setting as output drops/OutDiscards measures; this is due to the reasons above.
<Fast Facts>
We are often asked about degradation of traffic forwarding performance due to increase in the Total output drops/OutDiscards counter caused by configuring QoS.
Many cases will improve by reviewing the QoS configuration while bearing in mind the below three points:
- Impact of increase in Queue
- Impact of buffer that can be used in Queue being limited
- Impact of 1/25 shaping of Queue 1
The below is the setting information of 15.0(2)SE. Also refer to Configuration Guide of each IOS version.
Configuring QoS
Related Information
Original Document: https://supportforums.cisco.com/ja/document/12335411
Author: Yasuhiro Nakajima
Posted on Oct 24, 2014
Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community: