Comments on: Bridging the gap between 3550 and 3560 QoS: Part I http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/ Helping you become a Cisco Certified Internetwork Expert Tue, 07 Feb 2012 02:21:38 +0000 hourly 1 http://wordpress.org/?v=3.2.1 By: CiscoSri http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-145521 CiscoSri Fri, 08 Oct 2010 03:43:40 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-145521 Hi Petr, You article is really fantastic. I have a basic doubt, I am following the QoS SRND recommendation and have created the below policies and appied to the Access Port 0/5 & 0/9 respectively. I tested the policy which I have a class ftpdata and infact it does affect teh transfer rate however what is really botheing me is why am I unable to see any values in the below output on a Cisco 3560 Switch. Could you kindly explain why ? class-map match-all ipcommunicator match access-group name ipcommunicator class-map match-all http match access-group name http class-map match-all sccp-signaling match access-group name sccp-signaling class-map match-all ftpdata match access-group name ftpdata ! ! policy-map testpolicy class ipcommunicator trust dscp class ftpdata police 8000 8000 exceed-action drop class http police 64000 8000 exceed-action drop class class-default policy-map sccp-signaling class sccp-signaling police 32000 8000 exceed-action policed-dscp-transmit trust cos ! PHONESW#show policy-map interface fastEthernet 0/9 FastEthernet0/9 Service-policy input: sccp-signaling Class-map: sccp-signaling (match-all) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: access-group name sccp-signaling Class-map: class-default (match-any) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: any 0 packets, 0 bytes 5 minute rate 0 bps ! PHONESW#show policy-map interface fastEthernet 0/5 FastEthernet0/5 Service-policy input: testpolicy Class-map: ipcommunicator (match-all) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: access-group name ipcommunicator Class-map: ftpdata (match-all) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: access-group name ftpdata Class-map: http (match-all) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: access-group name http Class-map: class-default (match-any) 0 packets, 0 bytes 5 minute offered rate 0 bps, drop rate 0 bps Match: any 0 packets, 0 bytes 5 minute rate 0 bps ! Regards, CiscoSri Hi Petr,

You article is really fantastic. I have a basic doubt, I am following the QoS SRND recommendation and have created the below policies and appied to the Access Port 0/5 & 0/9 respectively. I tested the policy which I have a class ftpdata and infact it does affect teh transfer rate however what is really botheing me is why am I unable to see any values in the below output on a Cisco 3560 Switch. Could you kindly explain why ?
class-map match-all ipcommunicator
match access-group name ipcommunicator
class-map match-all http
match access-group name http
class-map match-all sccp-signaling
match access-group name sccp-signaling
class-map match-all ftpdata
match access-group name ftpdata
!
!
policy-map testpolicy
class ipcommunicator
trust dscp
class ftpdata
police 8000 8000 exceed-action drop
class http
police 64000 8000 exceed-action drop
class class-default
policy-map sccp-signaling
class sccp-signaling
police 32000 8000 exceed-action policed-dscp-transmit
trust cos

!
PHONESW#show policy-map interface fastEthernet 0/9
FastEthernet0/9

Service-policy input: sccp-signaling

Class-map: sccp-signaling (match-all)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: access-group name sccp-signaling

Class-map: class-default (match-any)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: any
0 packets, 0 bytes
5 minute rate 0 bps
!
PHONESW#show policy-map interface fastEthernet 0/5
FastEthernet0/5

Service-policy input: testpolicy

Class-map: ipcommunicator (match-all)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: access-group name ipcommunicator

Class-map: ftpdata (match-all)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: access-group name ftpdata

Class-map: http (match-all)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: access-group name http

Class-map: class-default (match-any)
0 packets, 0 bytes
5 minute offered rate 0 bps, drop rate 0 bps
Match: any
0 packets, 0 bytes
5 minute rate 0 bps
!
Regards,
CiscoSri

]]> By: Mohammed http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-126863 Mohammed Sat, 07 Aug 2010 22:00:10 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-126863 Hi Petr if we enable priority-queue on the interface, it will starve the other queues. is there any way to enable prioirty-queue and limit the bandwidth at the same time?? lets say i want to have priority-queue and guarantee it 20% of the bandwidth Also, is it possible to combine the ingress and output queues to guarantee(and limit) bandwidth for the priority queue? Regards Hi Petr

if we enable priority-queue on the interface, it will starve the other queues. is there any way to enable prioirty-queue and limit the bandwidth at the same time??

lets say i want to have priority-queue and guarantee it 20% of the bandwidth

Also, is it possible to combine the ingress and output queues to guarantee(and limit) bandwidth for the priority queue?

Regards

]]> By: jeremy http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-125294 jeremy Mon, 02 Aug 2010 11:18:49 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-125294 Hi Petr, What are drop-threshold1 and 2 means? I haven't found an explanation anywhere. max threshold as you explained is understandable but still confused what are drop-Threshold 1 and drop-threshold 2 means. Hi Petr,

What are drop-threshold1 and 2 means? I haven’t found an explanation anywhere.

max threshold as you explained is understandable but still confused what are drop-Threshold 1 and drop-threshold 2 means.

]]> By: Andy M http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-73978 Andy M Thu, 22 Oct 2009 16:03:21 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-73978 And of course I speaketh rubbish the bandwidth for shaped queue 1 would be 1/25th of the bandwidth and not 25/25. The remaining 3 queues can have a ratio of 25:25:25 or a 3rd of what is left. So 1/25 = 4% of the bandwith for Q1, the remaining queues share the 96 remaining percent which would be 32% each. So until I change my mind? I love it when you submit something they you think what rubbish I speak. And of course I speaketh rubbish the bandwidth for shaped queue 1 would be 1/25th of the bandwidth and not 25/25. The remaining 3 queues can have a ratio of 25:25:25 or a 3rd of what is left. So 1/25 = 4% of the bandwith for Q1, the remaining queues share the 96 remaining percent which would be 32% each.

So until I change my mind?
I love it when you submit something they you think what rubbish I speak.

]]> By: Andy M http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-73956 Andy M Thu, 22 Oct 2009 15:38:43 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-73956 Your question is not that simple to answer because we do not know what queues your traffic is destined for, however I think Q2,3,4 are not shaped and Petr’s article says “1/weight*{interface speed} gives you the bandwidth. You would have 25/25 *{interface speed} or basically the full interface being the shape limit for Q1. You would however have 25% of the interface available for each queue so the answer would depend on how much traffic was destined for which queue and if priority queuing had been enabled. If it was exactly 25% destined for each queue then all traffic would get an even distribution. If there was no traffic for Q1 then you would have 33% of the bandwidth for Q2,3,4. Enabling QOS “mls qos” and leaving the default settings can cause degradation in performance. Cisco assigned a bug to this at some point hence using auto QOS as a starting point is your best bet. Queue-sets 1 and 2 are aimed at different interface speeds but Cisco contradicts themselves as to what these are. There is no one size fits all for a QOS policy but autoqos is a great start for most networks. I agree about Cisco’s contradiction in the priority-queue stakes, in that you cannot shape Q1 if PQ is enabled. Indeed I tested this by throwing 30MB traffic at a 10MB interface of which 95% was EF traffic the remaining 5% was DSCP 0 – no DSCP 0 traffic got through even though I configured shaping, if you remove PQ then the shaping did as it was told. The other big issue with configuring these buffer space partitioning parameters as detailed by Petr and Cisco agree don’t touch unless you need to. Unfortunately the environment you need to configure these will be slow speed interfaces such as a 10MB MAN link. Here you need to buffer traffic as multiple 100 and even Gig links get pushed up a 10MB link. The 3560/3750 have port ASICS that controls a group of ports, this ASIC is basically the QOS buffering and scheduling ASIC, and how many ethernet ports share an asic depends on what model they are, some are 4 ports/asic some are 26. The buffer on the ASIC is a not particularly large but is a “Cisco confidential” amount of buffer. Unfortunately this also means that whilst one port on the asic is buffering for a slow link that buffer is also not available for an adjacent port, and can under certain circumstances degrade the performance of that adjacent port. You also mention your switch allows maximum-threshold up to 3200%, this value was increased in later versions of code from 400% to 3200%. I needed to deploy a config beyond the 400% limit and upgraded to one of these later codes to allow me to deploy this and greatly improved the performance of a converged network. Your question is not that simple to answer because we do not know what queues your traffic is destined for, however I think Q2,3,4 are not shaped and Petr’s article says “1/weight*{interface speed} gives you the bandwidth. You would have 25/25 *{interface speed} or basically the full interface being the shape limit for Q1. You would however have 25% of the interface available for each queue so the answer would depend on how much traffic was destined for which queue and if priority queuing had been enabled. If it was exactly 25% destined for each queue then all traffic would get an even distribution. If there was no traffic for Q1 then you would have 33% of the bandwidth for Q2,3,4.

Enabling QOS “mls qos” and leaving the default settings can cause degradation in performance. Cisco assigned a bug to this at some point hence using auto QOS as a starting point is your best bet. Queue-sets 1 and 2 are aimed at different interface speeds but Cisco contradicts themselves as to what these are. There is no one size fits all for a QOS policy but autoqos is a great start for most networks.

I agree about Cisco’s contradiction in the priority-queue stakes, in that you cannot shape Q1 if PQ is enabled. Indeed I tested this by throwing 30MB traffic at a 10MB interface of which 95% was EF traffic the remaining 5% was DSCP 0 – no DSCP 0 traffic got through even though I configured shaping, if you remove PQ then the shaping did as it was told.

The other big issue with configuring these buffer space partitioning parameters as detailed by Petr and Cisco agree don’t touch unless you need to. Unfortunately the environment you need to configure these will be slow speed interfaces such as a 10MB MAN link. Here you need to buffer traffic as multiple 100 and even Gig links get pushed up a 10MB link. The 3560/3750 have port ASICS that controls a group of ports, this ASIC is basically the QOS buffering and scheduling ASIC, and how many ethernet ports share an asic depends on what model they are, some are 4 ports/asic some are 26. The buffer on the ASIC is a not particularly large but is a “Cisco confidential” amount of buffer. Unfortunately this also means that whilst one port on the asic is buffering for a slow link that buffer is also not available for an adjacent port, and can under certain circumstances degrade the performance of that adjacent port.

You also mention your switch allows maximum-threshold up to 3200%, this value was increased in later versions of code from 400% to 3200%. I needed to deploy a config beyond the 400% limit and upgraded to one of these later codes to allow me to deploy this and greatly improved the performance of a converged network.

]]> By: gnijs http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-68958 gnijs Sat, 03 Oct 2009 12:58:53 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-68958 First, very good article. I am trying to figure out the details myself for over 5 days now :-) I do have some additional question that i am unable to solve myself: 1) with default auto-qos, when is it recommended to use Queue-set 1 or queue-set 2 ?? (note: we just had big trouble in our production network with using queue-set 2) 2) Using the default settings of mls qos: srr-queue bandwidth shape 25 0 0 0 srr-queue bandwidth share 25 25 25 25, in a full congested situation, how much bandwidth is assigned to Q2,3 and 4 ? Q1 gets 1/25th so 4%. I guess Q2,3,4 gets the remaining bits eqaully devided, but this is not 25% but 96% / 3 = 32% 3) The maximum-threshold value specifies how much a queue is allowed to grow and can be up to 400% (although my switch allows me to set it up to 3200 (??). Anyway, this 400% does it apply to the reserved buffer size ? mls qos queue-set output 1 buffers 10 10 70 10 Q3 7-times larger than other queues. reservation 50% each queue. Supposse i use 400% growth for every buffer. Does it mean Q1 can grow to maximum 40% of available buffer space in absense of other traffic ? If i want to allow each queue to use 100% of available buffer space in absence of other traffic, to i need to specify: 1000% 1000% 142% 1000% as growth rates ? First, very good article. I am trying to figure out the details myself for over 5 days now :-)
I do have some additional question that i am unable to solve myself:
1) with default auto-qos, when is it recommended to use Queue-set 1 or queue-set 2 ?? (note: we just had big trouble in our production network with using queue-set 2)

2) Using the default settings of mls qos:
srr-queue bandwidth shape 25 0 0 0
srr-queue bandwidth share 25 25 25 25,
in a full congested situation, how much bandwidth is assigned to Q2,3 and 4 ?
Q1 gets 1/25th so 4%. I guess Q2,3,4 gets the remaining bits eqaully devided, but this is not 25% but 96% / 3 = 32%

3) The maximum-threshold value specifies how much a queue is allowed to grow and can be up to 400% (although my switch allows me to set it up to 3200 (??). Anyway, this 400% does it apply to the reserved buffer size ?
mls qos queue-set output 1 buffers 10 10 70 10
Q3 7-times larger than other queues. reservation 50% each queue. Supposse i use 400% growth for every buffer. Does it mean Q1 can grow to maximum 40% of available buffer space in absense of other traffic ? If i want to allow each queue to use 100% of available buffer space in absence of other traffic, to i need to specify: 1000% 1000% 142% 1000% as growth rates ?

]]> By: Maarten http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-16171 Maarten Thu, 18 Dec 2008 13:04:39 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-16171 Hi, We're in the middle of defining new QoS standards with srr-queuing. We have spended all 4 queues to serveral dscp values. In some queues we want to make differance in priority with buffer allocation and thresholds. For example we have in queue three dscp value 10, 18 and 26. To give 26 priority above 18 and 18 above 10 we decided to devide those with thresholds. Default every threshold in queue-set 1 gets 100% ofthe buffers queue 2 get 200%. Can anyone give his view of how the threshold drop setting should be with some explanation? Below are some more details DSCP 10: management traffic DSCP 18: bussinuess critical DSCP 26: interactive mls qos srr-queue output dscp-map queue 3 threshold 1 10 mls qos srr-queue output dscp-map queue 3 threshold 2 18 mls qos srr-queue output dscp-map queue 3 threshold 3 26 mls qos queue-set output 1 buffers 10 50 20 20 Thanks in advance, Maarten Hi,

We’re in the middle of defining new QoS standards with srr-queuing. We have spended all 4 queues to serveral dscp values. In some queues we want to make differance in priority with buffer allocation and thresholds. For example we have in queue three dscp value 10, 18 and 26. To give 26 priority above 18 and 18 above 10 we decided to devide those with thresholds. Default every threshold in queue-set 1 gets 100% ofthe buffers queue 2 get 200%.

Can anyone give his view of how the threshold drop setting should be with some explanation?

Below are some more details
DSCP 10: management traffic
DSCP 18: bussinuess critical
DSCP 26: interactive

mls qos srr-queue output dscp-map queue 3 threshold 1 10
mls qos srr-queue output dscp-map queue 3 threshold 2 18
mls qos srr-queue output dscp-map queue 3 threshold 3 26
mls qos queue-set output 1 buffers 10 50 20 20

Thanks in advance,

Maarten

]]> By: Steve http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-6620 Steve Sun, 24 Aug 2008 21:34:14 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-6620 Hi Petr, This article is great. You mention that the 3560 loses the internal DSCP method that the 3550 used. However, the DOCcd is still cluttered with references to it in various sections of the 3560 configuration guide. Is this an error on their part (copy and paste?) I'm referring to: http://www.cisco.com/en/US/docs/switches/lan/catalyst3560/software/release/12.2_44_se/configuration/guide/swqos.html Just search for 'internal dscp', as they mention it several times. Is this just the usual doc cd errors? Hi Petr,

This article is great. You mention that the 3560 loses the internal DSCP method that the 3550 used. However, the DOCcd is still cluttered with references to it in various sections of the 3560 configuration guide. Is this an error on their part (copy and paste?)

I’m referring to:
http://www.cisco.com/en/US/docs/switches/lan/catalyst3560/software/release/12.2_44_se/configuration/guide/swqos.html

Just search for ‘internal dscp’, as they mention it several times.

Is this just the usual doc cd errors?

]]> By: Nicholas Davitashvili http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-3958 Nicholas Davitashvili Wed, 09 Jul 2008 05:38:37 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-3958 Thank you so much for the quick response and all the information! Thank you so much for the quick response and all the information!

]]> By: Petr Lapukhov, CCIE #16379 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-3954 Petr Lapukhov, CCIE #16379 Wed, 09 Jul 2008 03:33:28 +0000 http://blog.ine.com/2008/03/03/bridging-the-gap-between-3550-and-3560-qos-part-i/#comment-3954 <i>Nicholas Davitashvili</i> The trick with SRR shape is that it "guarantees" and "limits" the bandwidth at the same time. That is, if you set shape weight to 20 on a 100Mbps interface, you will restrict the queue to 1/20*100=5Mbps but this will also be the guaranteed rate (i.e. if interface is congested, you wont get less than 5Mbps). The "share" weight is not taken in account when "shape" is specified. Nicholas Davitashvili

The trick with SRR shape is that it “guarantees” and “limits” the bandwidth at the same time. That is, if you set shape weight to 20 on a 100Mbps interface, you will restrict the queue to 1/20*100=5Mbps but this will also be the guaranteed rate (i.e. if interface is congested, you wont get less than 5Mbps). The “share” weight is not taken in account when “shape” is specified.

]]>