The QoS SRND does a very good job of explaining the queuing configuration of the most popular switch platforms. I will discuss these switch platforms in this blog and the references section of this documents will include the direct links to the switch related queuing configuration section.
The 2950 EI (enhanced image) switch supports QoS, while the 2950 SI (standard image) does not. If the 2950 switch includes Gigabit Ethernet uplinks, the switch is an EI switch and supports QoS. The 2950 EI image does not include QoS support. QoS support is integrated in the hardware (ASIC) of the switch. An IOS upgrade will not bring QoS support to this switch. The 2950 supports a 1P3Q1T output only queuing architecture in which QoS processing is turned on by default. The 2950 is the only switch I know of that has QoS turned on by default. The 2950 has a 4 transmit (output) queue architecture with no WRED support. The queue architecture is represented as 4Q1T (4 queues, one threshold to drop traffic). The 1T nomenclature is very confusing because there are 6500 series line card that represent 1T as one threshold which congestion avoidance (WRED) will use to determine when the queue will drop lower priority traffic. The 4Q1T architecture of the 2950 switch can be converted to a 1P3Q1T architecture by turning on the priority queue (PQ). The priority queue on the 2950 is turned on differently based on the version of native IOS on the router.
Older versions of auto QoS performed the following configuration on the 2950 switch:
wrr-queue bandwidth 20 1 80 0
no wrr-queue cos-map
wrr-queue cos-map 1 0 1 2 4
wrr-queue cos-map 3 3 6 7
wrr-queue cos-map 4 5
The wrr-queue bandwidth command can be confusing because it looks like a long string of numbers. Each number relates to the amount of resources the weighted round robin (WRR) scheduler will associate with each queue. The bandwidth is carved up as follows during congestion:
Queue 1 - 20%
Queue 2 – 1%
Queue 3 – 80%
Queue 4 – 0 turns on the priority queue.
The numbers that appear in this command are not percentages. 20 represents the numerator in a fraction, while the denominator is the sum of all values used for all queues. If the sum of all the queues equal 100, each variable will represent a percentage. The priority (expedite) queue is a strict priority scheduler. The priority queue will be serviced until it is empty. The default CoS to queue mappings covered in an earlier blog is turned off with the no wrr-queue cos-map command. Each wrr-queue cos-map command relates one or more markings to a queue. The first number used in the string of numbers is the queue number, while each subsequent number represents the CoS value. Notice that no traffic was mapped to queue 2 in early versions of auto QoS.
Cisco IOS 12.1(20)EA2 generates a different auto qos policy that aligns to the current QoS SRND recommendations.
wrr-queue bandwidth 10 20 70 1
no wrr-queue cos-map
wrr-queue cos-map 1 0 1
wrr-queue cos-map 2 2 4
wrr-queue cos-map 3 3 6 7
wrr-queue cos-map 4 5
Notice that the newer version of autoqos use queue 2 and the nomenclature to turn queue 4 into a priority queue has changed. The value of 1 makes the configuration of the 2950 consistent with the other Cisco switch platforms.
In the next blog, I will explain the various recommended Cisco markings and continue the switch congestion management (queuing) conversation.
REFERENCES
2950 and 2955 QoS Configuration Guide - 12.1(20)EA2
http://www.cisco.com/en/US/docs/switches/lan/catalyst2950/software/relea...
2950 and 2955 QoS Configuration Guide - 12.1(20)EA2
http://www.cisco.com/en/US/docs/switches/lan/catalyst2950/software/relea...
Dennis Hartmann, CCIE No. 15651, is a consultant with www.highpoint.com and author of Implementing Cisco Unified Communications Manager, Part 1. Dennis is also a lead instructor at Global Knowledge. Dennis has various certifications, including the Cisco CCVP, CCSI, CCNP, CCIP, and the Microsoft MCSE. Dennis has various specializations including unified communications, data center, routing & switching, service provider (MPLS and optical). Dennis has worked for various Fortune 500 companies, including AT&T, Sprint, Merrill Lynch, KPMG, and Cabletron Systems. He lives with his wife and children in Hopewell Junction, New York.
The Leader in Network Knowledge
time to leave
Time to leave this blog, I am tired of reading boring copy and pastes from Global Knowledge and Cisco.
A blog it suppose to get your readers to want more, not to copy and paste technical information week to week.
Dear Cisco Subnet,
please find another blogger and pull Hartman, he is a waste of space on this web page. His blog name is Cisco UC, tell him to write about Cisco UC and not a technical guide on how to do QOS. Write about Cisco UC, new systems, phones and other things, thats a blog. There are 5 other bloggers on the subnet that have written more about Cisco UC than Hartman. Thats sad.
A former reader of his blog.
I find Mr. Hartmann's blogs
I find Mr. Hartmann's blogs quite informative and well-written. It seems the "former reader of this blog" is good at being demeaning, not constructive. Keep up the good work Hartmann!
Thanks
Thank you. I appreciate everyone's positive comments and technical discussions or questions.
Switch QOS
Right now, we are adding QOS to all of our switches in preparation for Cisco phone implementation. We have a variety of switches, most of which have been converted to Layer3 uplinks, 3560, 3750, 4506/4507, and 6513. So I am watching this blog with extreme interest to see if I have missed anything in the way that we configured QOS on our switches (mostly with auto QOS). Even though using auto QOS is pretty standard in its execution, it seems like the resulting macros on each platform have a different queue and syntax for the priority queue.
Auto QoS and switches
Auto QoS on the switches is a good start, but you probably want to tune the policy that Auto QoS generates to meet your needs. One of my customers used SAP and wanted to prioritize all traffic to and from the SAP servers. I used a classification and marking policy at each client's access port to mark the traffic going to the SAP servers as mission critical (AF31). I also had a policy on each of the two server ports that marked all traffic coming from these ports as AF31. My preference is to always identify traffic by port number if possible, but the customer did not have the port numbers and we didn't have the time to run a sniffer on the port and identify the traffic.
Hartman
I found this recent blog condensed and informative as we are actually looking into a 2950 volume refurb purchase with Qos a hot topic.
Thanks Dennis
2950
Thanks. I would not recommend the 2950 due to the TCAM resource limitations. It's very nice to be able to do classification and marking on port ranges (especially for voice media). Check out the 3550 if you're looking for a deal. The 3550 does not support output policing, but it's less limited than the 2950. The 3550 will definitely cost more, but it's probably worth it. The 2950 supports policing in 1Mbps increments, but does NOT support re-marking exceeding traffic into the scavenger class. Good luck and thanks for reading.
funny
Really funny how all of the sudden Hartman has people coming to his blog when someone said they are leaving. He must have asked a few friends to post for him. lol
Thanks for the time and
Thanks for the time and effort you put on these blogs Dennis. They are very informative and straight to the point. Keep up the good work.
Great Info, but not to funny!
Great info and an eye-opener on technology that I need to uo to speed on. Anon your post said you were leaving, what are you still doing here. Morbird curiosity or just angry that other people are not posting hate and discontent?
Post new comment