ASR 9000 Modular QoS Bundle link

jigojar · ‎09-16-2018

Hi Genius,

As of my understanding. If we have configured the Qos on virtual interface of bundle then policy will be applied on the each member interface.

In my case we have configured the QoS aggregate mode in virtual interface of bundle link to overcome to allocate the procured bandwidth on each member so that customer will not get the double the procured bandwidth.

but the problem here is if customer trying to push the traffic to single destination then i am getting only half of the bandwidth because in aggregate mode bandwidth is shared to both the physical interface. is their any solution to overcome this problem?

tkarnani · ‎09-17-2018

are you using sub interfaces? if so can try shared policy instance

https://www.cisco.com/c/en/us/td/docs/routers/asr9000/software/asr9k_r4-1/qos/configuration/guide/qc41asr9kbook/qc41clas.html#pgfId-1895131

jigojar · ‎09-17-2018

Thanks,

yes, actually we have configured the subinterface and applied the policy. here the problem is we have configured 10MB policy restriction in the subinterface and customer trying to use the 10MB to to the same destination but customer is getting only 5 MB for that particular destination.

As per my understanding 10MB policy is shared with the two physical members 5MB+5MB. here what i am looking i want to load share the traffic on the both the member interface even though if traffic goes to same destination. Pls confirm is their any workaround or solution to iur requirement

xr-escalation · ‎09-18-2018

Can you explain what do you mean by "same destination"? Are you referring to a single flow where the source IP address, destination IP address, source UDP/TCP port and destination UDP/TCP port are the same? If that is your use case, there are no ways to force that flow across more than bundle member because the hash calculation result will always be the same. So to allow that flow to achieve 10Mbps, you need to disable the "hw-module all qos-modebundle-qos-aggregate-mode".

Outside of a lab, a design scenario where there would be only one flow crossing the router is extremely unlikely. So in production you may fare better with aggregate mode because the hash calculation results will give enough variety to spread the load across two members.

You could also explore pseudo MC-LAG, where you load balance per VLAN. So all traffic of a single BE sub-interface (presuming you have two members) would hit always only one bundle member.

/Aleksandar

jigojar · ‎09-18-2018

Hi Alek,

Thanks, i am referring the same destination is destination-ip traffic in qos aggregate mode. if i am disabling the Qos aggregate mode then policy applied on the each member interface and customer can enjoy with the double the bandwidth. for example in normal qos mode if we configured 10MB policy restriction on the bundled sub interface then customer can able to use the 20MB traffic for multiple destination-ip and 10MB traffic for single destination-ip. If we configured the qos-aggregate mode then customer can use the 10MB for multiple destination & 5MB for single destination-ip. but our requirement is customer should use only 10MB either multiple destination traffic or single destination traffic. Pls confirm do we have any option?

Aleksandar Vidakovic · ‎09-19-2018

With single destination IP you still have three more parameters that are used to calculated the load-balancing hash: source IP, source L4 port, destination L4 port. So even in the aggregated mode traffic to a single destination IP will take multiple bundle members.

Aleksandar Vidakovic · ‎09-17-2018

What kind of traffic are you sending? If it's IPv4/IPv6, we should load-balance on a 5-touple (src/dst IP, src/dst port, router ID), so you should get a good spread.

If all traffic of a customer is destined to a single IP address (because of NAT), you could make use of "bundle load-balancing hash dst-ip" to force all the traffic to a single bundle member.

jigojar · ‎09-17-2018

Hi,

i have applied the command even though i can able to use only half of the configured bandwidth.

pls find the logs

===============

interface Bundle-Ether8.21
description SERVICE-POLICY-TEST
service-policy input RATELIMIT-IN-20MB
service-policy output RATELIMIT-OUT-20MB
ipv4 address 10.11.72.1 255.255.255.252
ipv4 address 10.11.81.1 255.255.255.252 secondary
ipv4 address 10.11.91.1 255.255.255.252 secondary
bundle load-balancing hash dst-ip
load-interval 30
encapsulation dot1q 21
!

RP/0/RSP0/CPU0:TN-LAB-AIR-T1-ST-MCN-MRT-CA91-R2#show policy-
policy-lib policy-map
RP/0/RSP0/CPU0:TN-LAB-AIR-T1-ST-MCN-MRT-CA91-R2#show policy-map interface Bundle-Ether8.21
Tue Sep 18 11:05:13.241 IST

Bundle-Ether8.21 input: RATELIMIT-IN-20MB

Class class-default
Classification statistics (packets/bytes) (rate - kbps)
Matched : 28376452/42564678000 19751
Transmitted : N/A
Total Dropped : 13948114/20922171000 9503
Policing statistics (packets/bytes) (rate - kbps)
Policed(conform) : 14428338/21642507000 10248
Policed(exceed) : 13948114/20922171000 9503
Policed(violate) : 0/0 0
Policed and dropped : 13948114/20922171000

Bundle-Ether8.21 output: RATELIMIT-OUT-20MB

Class class-default
Classification statistics (packets/bytes) (rate - kbps)
Matched : 14463209/21694813500 10248
Transmitted : 14419755/21629632500 10885
Total Dropped : 43869/65803500 0
Queueing statistics
Queue ID : 65578
High watermark : N/A
Inst-queue-len (packets) : 0
Avg-queue-len : N/A
Taildropped(packets/bytes) : 43869/65803500
Queue(conform) : 14419755/21629632500 10885
Queue(exceed) : 0/0 0
RED random drops(packets/bytes) : 0/0

RP/0/RSP0/CPU0:TN-LAB-AIR-T1-ST-MCN-MRT-CA91-R2#

jigojar · ‎09-17-2018

Hi,

Pls find the load-balance logs here.

RP/0/RSP0/CPU0:TN-LAB-AIR-T1-ST-MCN-MRT-CA91-R2#show bundle load-balancing bundle-ether8.21 location 0/RSP0/CPU0
Tue Sep 18 11:06:28.495 IST

Bundle-Ether8
Type: Ether (L3)
Members <current/max>: 2/64
Total Weighting: 2
Load balance: Default
Locality threshold: 65
Avoid rebalancing? False
Sub-interfaces: 1

Member Information:
Port: LON ULID BW
-------------------- --- ---- --
Te0/0/0/1 0 0 1
Te0/0/2/3 1 1 1

Sub-interface Information:
Sub-interface Type Load Balance Locality
Hash Threshold
---------------------------- ---- ------------ ---------
Bundle-Ether8.21 L3 Dst IP 65

===============================================================================
RP/0/RSP0/CPU0:TN-LAB-AIR-T1-ST-MCN-MRT-CA91-R2#

xr-escalation · ‎09-18-2018

The idea behind the "bundle load-balancing hash dst-ip" is to force all traffic towards a single destination IP address onto the same bundle member. So it shouldn't be used in conjunction with the aggregate QoS mode. This command is a perfect fit for BNG because a single customer has a single IP address.

I'm not sure what's your deployment scenario, but please do take a look into the BRKSPG-2904 from Cisco Live US or EUR from 2017. There's a long section on load-balancing.

In L3 deployment scenarios there is typically enough variation in the src/dst ip/port to allow for good load balancing. In L2VPN scenario we do have tweaks (explained in the Cisco Live presentation) that will allow you to achieve good load-balancing inside of a PW.

/Aleksandar