QoS issues with ASR 1001

[patrickdonlon]

Hi All

I've an ASR1001 with 15.1(2)S code on it connected to out ISP, we've been get some complaints about performance and I'm seeing drop on the output policy. Checking the bandwidth consumption we have plenty spare when drops are occuring, there's 300Mb/s.  Details below, any suggestions gratefully received

Thanks

Patrick

The policy is to guarantee the following bandwidth:

Outbound policy:

class 1 => 50% guaranteed

class 2 => 8% guaranteed

class 3 => 1% guaranteed

class 4 => 5% guaranteed

class 5 => 1% guaranteed

class 6 => 5% guaranteed

class  => 7% guaranteed

class 8 => 7% guaranteed

class default => not configured

config :

policy-map priority

class 1

  priority percent 50

class 2

  priority percent 8

class 3

  priority percent 1

class 4

  priority percent 5

class 5

  priority percent 1

class 6

  priority percent 5 2000000

class 7

  priority percent 7

class 8

  priority percent 7

class class-default

policy-map outbound-internet

class class-default

  shape average 300000000

  service-policy priority

and the drops are seen :

Service-policy output: outbound-internet

Class-map: class-default (match-any)

157386235086 packets, 142277415785757 bytes

30 second offered rate 158203000 bps, drop rate 0000 bps

Match: any

Queueing

queue limit 1250 packets

(queue depth/total drops/no-buffer drops) 0/83753649/0

(pkts output/bytes output) 157302481432/142155881947373

shape (average) cir 300000000, bc 1200000, be 1200000

target shape rate 300000000

Service-policy : priority

queue stats for all priority classes:

Queueing

queue limit 512 packets

(queue depth/total drops/no-buffer drops) 0/83753649/0

(pkts output/bytes output) 149622194962/136919622255672

Class-map: 1 (match-all)

38746601707 packets, 44785520236642 bytes

30 second offered rate 47228000 bps, drop rate 0000 bps

Match: access-group name web-farm

Priority: 50% (150000 kbps), burst bytes 3750000, b/w exceed drops: 22745

Class-map: 2 (match-all)

32261813730 packets, 23350595823948 bytes

30 second offered rate 11610000 bps, drop rate 0000 bps

Match: access-group name webfarm-VPN

Priority: 8% (24000 kbps), burst bytes 600000, b/w exceed drops: 3959031

Class-map: 3 (match-all)

258399421 packets, 267198408224 bytes

30 second offered rate 0000 bps, drop rate 0000 bps

Match: access-group name web-farm-mail

Priority: 1% (3000 kbps), burst bytes 75000, b/w exceed drops: 87883

Class-map: 4 (match-all)

17160556172 packets, 4209151495211 bytes

30 second offered rate 18982000 bps, drop rate 0000 bps

Match: access-group name corporate-browsing

Priority: 5% (15000 kbps), burst bytes 375000, b/w exceed drops: 129850

Class-map: 5 (match-all)

239923 packets, 42322375 bytes

30 second offered rate 0000 bps, drop rate 0000 bps

Match: access-group name newscientist-media

Priority: 1% (3000 kbps), burst bytes 75000, b/w exceed drops: 0

Class-map: 6 (match-all)

34710766533 packets, 49678200430642 bytes

30 second offered rate 61048000 bps, drop rate 0000 bps

Match: access-group name corporate-DR-VPN

Priority: 5% (15000 kbps), burst bytes 2000000, b/w exceed drops: 79015405

Class-map: 7 (match-all)

14845163520 packets, 8016844416864 bytes

30 second offered rate 7632000 bps, drop rate 0000 bps

Match: access-group name corporate-VPN

Priority: 7% (21000 kbps), burst bytes 525000, b/w exceed drops: 40342

Class-map: 8 (match-all)

11721867776 packets, 6733254903938 bytes

30 second offered rate 10804000 bps, drop rate 0000 bps

Match: access-group name corporate

Priority: 7% (21000 kbps), burst bytes 525000, b/w exceed drops: 498393

Class-map: class-default (match-any)

7680247559 packets, 5236234911120 bytes

30 second offered rate 897000 bps, drop rate 0000 bps

Match: any

queue limit 1250 packets

(queue depth/total drops/no-buffer drops) 0/0/0

(pkts output/bytes output) 7680286470/5236259691701

[paul driver]

Hello

The queue-limits of the classes dont look like they are default? - I think I read ( NOT 100% certain) that the total numbers of class queue-limits cannot exceed the physical interfaces hold-queue limit? ( default 1000), if so tail drop occurs before the class queue limit is reached .

Maybe someone else can jump in here to clarify this?

Regards your qos config, you could  try not performing LLQ on all classless other than low latency traffic such as voice/video.

eg:

Class 1
Priority xx
Class 2
Bandwidth percent xx
etc......
Class class-default

Policy-map outbound-Internet
Class class-default
Shape 3000000000
Policy priority

Also
sh policy-map int xxxx

sh queue xxx(int)

res

Paul

Please don't forget to rate any posts that have been helpful.

Thanks.

[patrickdonlon]

Thanks for this, I was a little curious about the LLQ on all the queues as we don't have any voice or video on this.

How do I check the class queue limits vs the interface hold queue?

Thanks

[paul driver]

Hello

Can you post your config

and also

sh policy-map int xxxx

sh queue xxx(int)

res

Paul

Please don't forget to rate any posts that have been helpful.

Thanks.

[patrickdonlon]

Hi Paul

Config and sh policy map already posted, have I missed something? The show queue isn't support on this box, it's IOS-XE, 15.1.

Thanks

[paul driver]

Hello Patrick,

I can see an aggregate queue limit of 512 which is for all 8 classes

8*64= 512 (64 packets is the default for a traffic class) 

Doe any of your physical interfaces have the hold-queue xx in/out  or queue-limit xx configured?

res

Paul

Please don't forget to rate any posts that have been helpful.

Thanks.

[patrickdonlon]

Hi Paul

No neither configured, can you explain the aggregate queue limit? I doing some reading on this subject now too

Thanks

[paul driver]

Hello Patrick

Going back to what I first stated regards the defining a queue-limit value on each traffic class in hierarchical qos, These values together ( aggregate) should not exceed the hold-queue limt ( default 1000) of the physical interface otherwise drops could occur.

http://www.cisco.com/en/US/prod/collateral/iosswrel/ps6537/ps6558/sol_ov_c22-708224.html

res

Paul

Please don't forget to rate any posts that have been helpful.

Thanks.

[patrickdonlon]

Hi Paul

Was also reading this doc it's very good. So if I have 8 low latency queues do these share the 512 limit?

Looks like I need to redefine the queues as non LLQs as a starting point anyway

Regards

Patrick

[paul driver]

Hello Patrick,

Was also reading this doc it's very good. So if I have 8 low latency queues do these share the 512 limit? - Yes they total to 512 packets -thats what we can see in your output

Try and not to LLQ if you dont need to and apply a BW percentage to each class and maybe even apply WRED to them also for congestion avoidance

Class 1

Bandwidth percent xx

random-detect

Class 2

etc..

Class 3

etc..

res

Paul

Please don't forget to rate any posts that have been helpful.

Thanks.