QOS CBWFQ Question/Issue

[adam.stlaurent]

hello

I am hoping someone can help me understand this concept.

I configured Class based weighted fair queuing seen below to give priority to citrix return traffic coming from 2 particular subnets (1.1.1.1 and 192.168.100.0 in my example to the customers site). Now the circuit size is 4.5 mb and it does get saturated so the QOS map starts going into affect. now suppose the citrix traffic only accounts for 800kbps on the circuit but we gave it guaranteed 85% of the 4.5mb. I understand everything else gets the last 15% but what happens to the extra bandwidth on the 85% policy since citrix is not actually using all of it. Is it wasted or is the class default allowed to go and take from bandwidth guaranteed to something else. Also is there anything else i am missing here, the customer keeps complaining about citrix and saying it is freezing on them. 

Thank you

interface Multilink3

description 3 Bonded T1's

ip vrf forwarding VRF

ip address 192.168.1.1 255.255.255.252

ip flow ingress

max-reserved-bandwidth 100

service-policy output PMAP

no cdp enable

ppp multilink

no ppp multilink fragmentation

multilink-group 3

no clns route-cache

end

policy-map PMAP

  class QOS

   set ip precedence 5

   bandwidth percent 85

class-map match-any QOS

  match access-group 155

show access-list 155

Extended IP access list 155 (Compiled)

    10 deny icmp any any (1007428 matches)

    20 permit ip host 1.1.1.1 172.16.2.0 0.0.0.255 (23872569 matches)

    30 permit ip 192.168.100.0 0.0.0.255 172.16.2.0 0.0.0.255 (14854804 matches)

[David_Che]

If total traffic qos + default < = 4.5 M bps, neither traffic will be dropped.

if total traffic qos + default > 4.5 M bps and qos traffic <= 4.5 * 85%, only drop default traffic

if both qos > 4.5 *85% and default traffic > 4.5 *15%, both traffic will be dropped and QOS will be guaranteed with 85% bandwidth size.

You can tried to mark only CITRIX application into class QOS as below and see the outcome. 

access-list 155

 deny icmp any any 

 permit tcp host 1.1.1.1 172.16.2.0 0.0.0.255 equal citrix

 permit tcp 192.168.100.0 0.0.0.255 172.16.2.0 0.0.0.255 equal citrix

HTH

[adam.stlaurent]

Ok that is making a little bit more sense. citrix is the only thing at those subnets so i do not need to match the protocol sense it is the only thing there. 

I am still trying to understand

if Qos < 85% of 4.5 mb for example it is only using for example 45% and default traffic is high so the policy goes into effect. so the default traffic get throttled and dropped down to 15% but what happens to the other 40% that is in the QoS poliy. Since citrix is actually only using 45% of the 85% guarenteed is the other 40% just lost. 

so basically a 4.5mb circuit would only be partially used and waiting if citrix traffic wanted to grow to 85%?

[David_Che]

the left QOS 40% bandwidth will be occupied by default traffic, in other words, QOS occupied 45% bandwidth, default 55% bandwidth.

if QOS traffic goes up to 70%, then default traffic will occupy the left 30%. 

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Generally, "extra" available bandwidth is made available to CBWFQ queues, how it's proportioned, can vary based on device/IOS, but often it's made available to the classes based on their bandwidth ratios.

BTW, there are some major changes in CBWFQ between pre-HQF and HQF.  Knowing the platform and IOS might help.  (With compiled ACLs, this a 7200 platform?)

Sometimes you need to decrease interface tx-rings sizes to cause interface queued packets to be processed by CBWFQ rather than interface's tx-ring FIFO.

If your IOS is providing pre-HQF, I don't recall whether an implicit class-default uses FIFO or WFQ.  It's often better to fully define the policy.

As to pauses in Citrix, you're sure your ACLs, which match IPs, only will "see" Citrix traffic?  Does the Citrix traffic support other than screen scraping traffic, e.g. Citrix printing or Citrix file copying, aren't being used.  (If the latter, some versions of NBAR can distinguish between Citrix traffic types for QoS purposes.)

[adam.stlaurent]

The ios is: Cisco IOS Software, 7200 Software (C7200-KBOOT-M), Version 12.4(3g), RELEASE SOFTWARE (fc2)

In general everything coming in from that subnet is citrix as well as printing over port 9100 from the citrix farm. They do some USB mapping in citrix as well for this like scanning and they save off some reports to local disk sharing as well from the citrix session. 

so basically we see this client as having 2 types of traffic. Everything heading from the citrix farm and internet. I understand if a print job happens it will be treated the same as citrix and if need be we can classify that separately but it looks like citrix does not even come close to the 85% reserved so i am hoping the extra is there for printing if needed.  

here is the show policy-map i change the class to 50% to try to even it out a bit as well.

show policy-map int mu3
Multilink3

Service-policy output: HCS

Class-map: HCS_QoS (match-any)
20354018 packets, 3913932699 bytes
5 minute offered rate 243000 bps, drop rate 0 bps
Match: access-group 155
20354018 packets, 3913932699 bytes
5 minute rate 243000 bps
QoS Set
ip precedence 5
Packets marked 20354018
Queueing
Output Queue: Conversation 265
Bandwidth 50 (%) Max Threshold 64 (packets)
(pkts matched/bytes matched) 20354018/3913932699
(depth/total drops/no-buffer drops) 0/3478/0

Class-map: class-default (match-any)
9149086 packets, 10553872403 bytes
5 minute offered rate 187000 bps, drop rate 0 bps
Match: any

[Joseph W. Doherty]

Disclaimer

The Author of this posting offers the information contained within this posting without consideration and with the reader's understanding that there's no implied or expressed suitability or fitness for any purpose. Information provided is for informational purposes only and should not be construed as rendering professional advice of any kind. Usage of this posting's information is solely at reader's own risk.

Liability Disclaimer

In no event shall Author be liable for any damages wha2tsoever (including, without limitation, damages for loss of use, data or profit) arising out of the use or inability to use the posting's information even if Author has been advised of the possibility of such damage.

Posting

Ok, your IOS version is pre-HQF.

Interesting that your Citrix class does show some drops.

What you might try is setting your serial interface's tx-ring-limit the minimal value and see what happens.

[Vasilii Mikhailovskii]

Hello.

Per statistics average packet size of the class is 192 bytes. Sounds like servers are on the opposite side and this QoS has no effect on server-to-client traffic flow.

"Citrix freezes" may come from packet loss on server->client direction (as normally bottleneck happens here).