Ask the Expert: QoS on Catalyst Switches. - Page 7

ciscomoderator · ‎05-18-2012

With Shashank Singh and

Welcome to the Cisco Support Community Ask the Expert conversation. This is an opportunity to learn from Cisco experts Shashank Singh and Sweta Morga about implementation and working and troubleshooting QoS on Cisco Catalyst 2960, 3650, 3750, 4500 and 6500 switches.

Shashank Singh graduated in 2009 with a bachelor's degree in Computer Science and Engineering from VIT University, Vellore India. Prior to joining Cisco he worked at General Electric as a software engineer. Later on he joined the Cisco Technical Assistance Center as an engineer in October of 2009. He has been working on LAN Switching technologies in TAC since then. Shashank also holds a CCNP certificate. QoS on Catalyst switches is one of the areas of his interest.

Sweta Mogra is a Computer Science & Engineering graduate from VIT University, India. She has worked as a consultant with Tata Consultancy Services before joining Cisco's Technical Assistance Center (TAC) in 2011. She is currently working on LAN Switching technologies and QoS as one of her areas of expertise.

Remember to use the rating system to let Shashank and Sweta know if you have received an adequate response.

Shashank and Sweta might not be able to answer each question due to the volume expected during this event. Remember that you can continue the conversation on the Network Infastructure sub-community Lan Switching forum shortly after the event. This event lasts through June 1, 2012. Visit this forum often to view responses to your questions and the questions of other community members.

bedapjain · ‎05-25-2012

I want to know q1, q2, q3 and q4 bandwidth for the following config

assume the interface bandwidth is 100M

Lets say shaped queue weight: 20 0 0 0

shared queue weight: 30 40 50 60

I think q1 will be 1/20 times 100 = 5M

q1 should not participate in sharing, so q1 will be fixed 5M rate limit

but q2 will 40/(405060) times 100 and that is not rate limit. please confirm it

what is the use of config 30 for q1 for sharing ?

what will be the queues bandwidth if egress priority queue is enable ?

Beda Jain | Consultant - CCIE #12267

13777 Ballantyne Corp Place | Suite 305 | Charlotte, NC 28277

Mobile: 9195220242

bjain@ineteng.com | http://www.ineteng.com<>

Shashank Singh · ‎05-25-2012

Hi Beda,

Following rules govern bandwidth calculation in sharing, shaping and priority queueing on 3750 platform:

If the egress priority queue is enabled, it overrides the SRR shaped and shared weights for queue 1.

If the egress priority queue is disabled and the SRR shaped and shared weights are configured, the shaped mode overrides the shared mode for queue 1, and SRR services this queue in shaped mode.

If the egress priority queue is disabled and the SRR shaped weights are not configured, SRR services this queue in shared mode.

Please find answers inline.

I think q1 will be 1/20 times 100 = 5M

Correct.

q1 should not participate in sharing, so q1 will be fixed 5M rate limit

Correct.

q2's share = 40/(30+40+50+60) % of 95M = 22M which is flexible depending on whether or not congestion is present.

For more details please check this thread - https://supportforums.cisco.com/thread/2055051

what is the use of config 30 for q1 for sharing ?

This weight does not affect bandwidth of q1 as it is overidden by the shaping weight.

what will be the queues bandwidth if egress priority queue is enable ?

If you use 'priority-queue out' command under an interface, as long as there is any traffic in the priority queue (queue 1 by default), 100% of the bandwidth is used by queue 1. Also, weight1 in the shape and share commands is not used in bandwidth calculation for other queues.

Regards,

Shashank

bedapjain · ‎05-25-2012

in this example egress priority queue is enabled what will be the bandwidth for q2, do we consider the q1 weigh for q2 calculation ?

Beda Jain | Consultant - CCIE #12267

13777 Ballantyne Corp Place | Suite 305 | Charlotte, NC 28277

Mobile: 9195220242

bjain@ineteng.com | http://www.ineteng.com<>

Shashank Singh · ‎05-25-2012

If priority queue is enabled, weight1 in the shape and share commands is not used in bandwidth calculation for other queues.

Regards,

Shashank

Shashank Singh · ‎05-28-2012

Hi Beda,

If priority queue is enabled, q1 weight is not used in weight calculation of any other queue.

Regards,

Shashank

stevetyler123 · ‎05-28-2012

Hi,

Any chance of MQC coming to the lower end platforms such as 3560 and 3750(X) ranges? When rolling our QoS policies across an Enterprise creating different configurations for different platforms is very time-consuming. Coupled with some of the bizarre behaviours of classification on the 3750 and 4500 switches it makes it a lot more difficult than it needs to be. Perhaps I can illustrate with a few examples :

1. 4500 Sup V TCAM Usage- Makes classification on ingress using ACL's impossible on a wiring closet switch as you run out of TCAM in a blink of an eye. Trying to calculate how much TCAM memory you're going to use is actually impossible due to the different results given depending on the order in which ACL's are applied etc. So we have to do Egress classification. Now it would be nice to have consistency across the estate but.....

2. 3750- Doesn't support classification by DSCP on egress- Why not?

3. 3750 and 4500 (Sup V)- Don't do MQC (The lovely 4500 Sup 7E/7LE does- hooray!)

4. Don't get me started on the 6500 where every module seems to have a different number of queues, thresholds and the priority queue changes between 1 and 4, or doesn't exist at all.

5. How about having some useful monitoring commands on the 3750 and 4500 (Sup V) range. Having to do a "sh platform etc etc etc etc" to work out whether you are dropping packets, and then wondering why the queues are numbered - 0 through 3 instead of 1 through 4, is really poor.

6. "show policy-map int x/y/z" on a 3750- doesn't tell you anything

7. "show policy-map int x/y/z" on a 4500 SupV- Shows an aggregation of the hits against ALL interfaces to which the policy-map on x/y/z is applied

I don't expect individual answers to these questions because that would be pointless. What I would like to see is some configuration and functionality uniformity (as Adam Styles said), added as an item on the product roadmaps. At the moment it's as though these products come from different companies.

Regards

Steve Tyler

Shashank Singh · ‎05-28-2012

Hi Steve,

Thanks for your valuable feedback!

The key challenge to implementing feaures on switches is that they need to be implemented in ASICs, which bumps up costs by huge margins. For the same reason we have MQC like flexible features implemented on core platforms like 6500/Sup7 but not on access layer platforms like 3750.

With advancements in ASIC design and them being available at a cheaper cost with time, there is a possibilty that QoS migrates to a unified approach. I am not sure whether this is there in the roadmaps yet but can definitely put your feedback across to correct teams.

Regards,

Shashank

Andrey.Miroshnichenko · ‎05-25-2012

Unfortunately "sh mls qos int gix/y stat" didn't display input drops too. It seems, that Cisco has not provided analyzing mechanism for this issue. Maybe, it is because input drops not always caused by bad QoS setting, as you explain.

But, this command is very helpful to monitor - correctly dscp marking policy works, or not.

Anyway, thank for your time!

glgersc · ‎05-25-2012

Is there any way to monitor the behavior and operation of a VLAN-based micro-flow policer on a 6500 switch? I'm using the Medianet 4.0 Design Guide as a reference. However, when trying to monitor the traffic counts, it appears that the micro-flow classes are counting every packet in the switch:

USSP-B1F1-SW01#sho mls qos ip vlan 98
Bridged QoS is enabled on Vlan98
[In] Policy map is QOS-EDGE-VLANBASED [Out] Default.
QoS Summary [IPv4]: (* - shared aggregates, Mod - switch module)

      Int Mod Dir Class-map DSCP Agg Trust Fl   AgForward-By   AgPoliced-By
                                   Id         Id
-------------------------------------------------------------------------------
      Vl98 5 In MATCH-VOIC    0    0* dscp 1    40661921325              0
      Vl98 5 In MATCH-VOIC    0    0* dscp 2    40661921325              0
      Vl98 5 In MATCH-VIDE    0    0* dscp 3    40661921325              0
      Vl98 5 In PRIBUSINES   26    8     No 0         246534              0
      Vl98 5 In BUSINESS-A   18    9     No 0        8808432              0
      Vl98 5 In SCAVENGER-   10   10     No 0              0              0
      Vl98 5 In class-defa    0   11     No 0        5716928              0

The non-micro-flow classes are counting and marking correctly. The micro-flow classes are at least being trusted, based on sniffer traces, but it's impossible to tell how many packets are in each class, or what, if anything the policers are doing. The 'sho policy-map int vlan 98' command agrees with the non-micro-flow traffic counts, but shows zero for the micro-flow classes.

Is there some other command that will give useful numbers for the micro-flow policer classes?

The VLAN interfaces in this switch do not have any L3 routing. They exist only as placeholders to apply the policy. For non-routing SVI interfaces, can you confirm whether the 'mls qos bridged' command is required?

Shashank Singh · ‎05-25-2012

What we are seeing is an expected behavior,show mls qos ip type mod/number command does not show the microflow policing statistics. It only shows the aggregate policing statistics. Please check this document

http://www.cisco.com/en/US/products/hw/switches/ps708/products_tech_note09186a008074d6b1.shtml

"show mls netflow ip detail nowrap" and "show mls ip detail "can be used to check all available stats (including per flow policed count) for microflow policing on Catalyst 6500 switches.

As you correctly understood "mls qos bridged" command is required if you want microflow policer to affect bridged traffic on that vlan.

Regards,

Shashank

glgersc · ‎05-25-2012

Thanks. I'm glad to hear that is a normal output. And thanks for the info on the additional commands.

Of course, more commands lead to more questions. The 'sho mls netflow ip detail nowrap' does not seem to show anything except multicast and broadcast traffic on the VLAN's that have the micro-flow policer policies applied. This is what I'm seeing on the voice VLAN while at least two calls are active:

USSP-B1F1-SW01#sho mls netflow ip detail nowrap | i Src|Vl98

DstIP SrcIP Prot:SrcPort:DstPort Src i/f

224.0.0.2 10.1.98.3 udp :1985 :1985 Vl98

239.1.1.2 10.1.98.4 udp :2052 :4097 Vl98

224.0.0.2 10.1.98.2 udp :1985 :1985 Vl98

The data VLAN is similar. I'm only seeing regular flows on the interfaces/VLAN's without 'mls qos vlan-based' or 'mls qos bridged'.

Are there specific 'mls flow' or interface/vlan 'netflow' commands that are also required?

mikejtay · ‎05-25-2012

Hi,

I have a two part question on 6500 LAN cards (65XX and 67XX) and sub-rate (sub-interface or vlan) queueing. The first has to do with the use and prioritization of the priority queue when there is NO congestion. Does the asic still empty priority queue packets prior to the bandwidth queues with no congestion or are all packets equal and FIFO in an uncongested situation? The second question is how does the asic determine there is congestion on LAN cards and can it be seen with a show command? I have never seen anything like a WAN card TX-ring mentioned anywhere. Thanks.

Shashank Singh · ‎05-25-2012

Hi,

Please find answers inline.

Does the asic still empty priority queue packets prior to the bandwidth queues with no congestion or are all packets equal and FIFO in an uncongested situation?

This is correct. On switches, priority queue is a strict priority queue, and is always serviced until empty before any other queue irrespective of congestion level.

How does the asic determine there is congestion on LAN cards and can it be seen with a show command?

Congestion is determined based on configured tail drop/WRED thresholds. These values depend on the line card being used and can be changed via configuration. To view current values "sh queueing interface gix/y" may be used.

Following document talks in detail about drop thresholds for various line cards:

http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper09186a0080131086.html

Regards,

Shashank

manumohan200 · ‎05-25-2012

Hi Shashank/Shweta,

Nowadays, you know that the Metro Ethernet links to connect locations within a city will be connected direclty to LAN switches like 3750 or 6500 etc.

Is there anyway to reserve specific bandwidth to certain traffic, say Voice (DSCP EF) in these switches?

For example, I have a Metro Ethernet link of 10 Mbps between Location A and Location B. I need to reserve 3 Mbps of bandwidth to Voice, 4 Mbps of bandwidth to Video. Consider it is a 6500 Switch with WS-6748 line card.

What would be the configuration?

Thanks,

Manu

Shashank Singh · ‎05-25-2012

Hi Manu,

We will need to find out as to which queue will voice and video go to. This will require us to know the dscp/cos markings of these traffic types and their association with queues through which the traffic is supposed to go out of.

dscp-queue association can be seen using "sh queueing interface gix/y" command.

Once association is known, We would need to assign bandwidth to these queues using the command "wrr-queue bandwidth precent X Y", where X and Y are percent of interface speed.

Assuming voice is mapped to queue 1, video to queue 2 and rest everything to q3, following command will help us acheive this on a 6748 LC:

switch(config-if)#wrr-queue bandwidth percent 3 4 93

Hope this helps.

Regards,

Shashank