10-04-2007 07:19 AM - edited 03-03-2019 07:01 PM
We recently saw this article:
http://www.cisco.com/en/US/tech/tk543/tk544/technologies_tech_note09186a0080094612.shtml
***snippet***
Understand Special Queues With Non-RSP Platform
As noted in the Packet Prioritization Tags and Queuing table, Cisco router platforms like the Cisco 7200, 3600 and 2600 series place pak_priority messages into a separate set of queues and not the class-default set of queues.
QoS and Locally Generated Packets
Traffic generated by the router represents a special case for outbound QoS service policies. Some locally generated traffic must be treated as any other user traffic, and the QoS system must apply the configured QoS mechanisms to this traffic. An example of such traffic is performance probes that are designed to measure the behavior incurred by packets of a given class. Other locally generated traffic, particularly Layer 2 keepalives and routing protocol messages, are vital to the basic functioning of the router and must not be subject to some QoS features. For example, weighted random early detection (WRED) must not drop Layer 2 keepalives when the average queue depth reaches a high watermark
In addition, packets destined to the router must be handled carefully. For example, remember that a service-policy that applies class-based policing must not apply to packets destined to the router to avoid dropping important control messages.
***end snippet***
If I read it correctly, we do not need a specific QOS policy for routing updates and control traffic.
The problem is, 2 years ago we rolled out the first QOS policy on 2651 routers. OSPF adjacencies were dropped during periods of congestion. One engineer finally marked OSPF traffic and put it in the priority queue, and the problems stopped.
This seems to contradict what the article was saying. We even had a Cisco rep on site for another project and we ran the question past him. He said that OSPF routing is treated differently and that we didn't need to include it in the QOS policy.
Does anyone have a good answer to this one? (thanks in advance)
CWB
10-04-2007 08:33 AM
Hi,
The routing protocol traffic is by default marked as cs6. But the same phenomenon i have seen it as well. when i did a lab emulation for the same. i could notice that when a bursty traffic affects a link QOS can misbehave. i guess , the same can be caused due to the queueing and dequeuing of different queues due to a bursty traffic.
I may be incorrect ok... but for the discussion sake i am sharing whatever i have done on the lab.
if you are mixing classes there things are going to happen. if you want every classes its unique bandwidth value ..i think you have to police each class so that it does not overflow to other classes. But it will be an ineffective method as your bandwidth will not be utilised properly.
Thanks and Regards,
Srinath.Muralidharan
Cable&Wireless
+919980930364
10-04-2007 08:54 AM
Thanks, Srinath.
I have some emails going on this topic and the responses vary quite a bit. From what I understand, the behavior of control traffic is dependent upon platform and even code version. For instance, on 7500-series or 6500s, you do have to make a policy because the pak_priority flag isn't treated the same as on 2600 and 7200 routers. More to come as I learn it...
CWB
10-07-2007 05:04 PM
Hi,
Please drop me a line if you find some documents on the same.
I also wish to join the learning party!!
Thanks and Regards,
Srinath.Muralidharan
Supp2 - Enterprise Network Management
Cable&Wirless
+919980930364
10-08-2007 07:10 AM
Haven't gotten definite confirmation on what interfaces the bandwidth statement absolutely affects QOS yet, but I did get confirmation that it is ignored on MLPPP links. We also confirmed that the bandwidth allocation is dynamic based on what links are in the bundle. Here is the output from when there are 2 links in the bundle, then a third link is added:
7206-rtr6#sh policy-map inter mu1
Multilink1
Service-policy output: def-bw
Class-map: voice (match-any)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: ip dscp ef (46)
0 packets, 0 bytes
5 minute rate 0 bps
Match: ip precedence 5
0 packets, 0 bytes
5 minute rate 0 bps
Queueing
Strict Priority
Output Queue: Conversation 264
Bandwidth 30 (%)
Bandwidth 921 (kbps) Burst 23025 (Bytes)
(pkts matched/bytes matched) 0/0
(total drops/bytes drops) 0/0
7206-rtr6#sh policy-map inter mu1
Multilink1
Service-policy output: def-bw
Class-map: voice (match-any)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: ip dscp ef (46)
0 packets, 0 bytes
5 minute rate 0 bps
Match: ip precedence 5
0 packets, 0 bytes
5 minute rate 0 bps
Queueing
Strict Priority
Output Queue: Conversation 264
Bandwidth 30 (%)
Bandwidth 1382 (kbps) Burst 34550 (Bytes)
(pkts matched/bytes matched) 0/0
(total drops/bytes drops) 0/0
I will update further if I get anything in regards to other interfaces and/or platforms...
CWB
Discover and save your favorite ideas. Come back to expert answers, step-by-step guides, recent topics, and more.
New here? Get started with these tips. How to use Community New member guide