cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1133
Views
0
Helpful
3
Replies

QoS implementation sugestions for SDWAN

Amar_Tufo
Level 1
Level 1

Hello all, 

I'm asking for bit of guadiance in QoS implementation for SDWAN, in a hub and spoke topology, where both have mpls and internet tloc's. Service side for hub and also spoke are in one VPN 10. On brach side the default VLAN1 is in the VPN10, on the hub side we have established BGP to OMP which advertises al the routes from BGP to OMP into VPN10. So the HUB is reachable through both tlocs.

In particular I seek for a way to implement QoS on the mpls interface, as we suffer from high bandwidth allocation to simple SMB file sharing folder browsing, where everything other is droped. adaptive Qos and shaping is enabled on the spoke mpls interfaces.

QoS per Tunnel is not possible as the documentation states "Per-tunnel QoS with loopback WAN for non-binding mode isn’t supported on the hub." and we use an loopack interface on the hub so we can agregate all the mpls links from different providers to a L3 switch to router. 

I've tried with Per-VPN QoS policy on spoke with an ACL, where it should refer to a QoSmap scheduler where a certain amount of bw is alocated to a specific traffic, but I don't see any improvement and no hit counts on the ACL. 

What are the community experiences?

3 Replies 3

Hi,

how does QoS policy look like in CLI and in vManage gui? Please, share it

HTH,
Please rate and mark as an accepted solution if you have found any of the information provided useful.

Thank you for reaching out. 

Our centralized policy does not treat QoS, it treats only VPN membership and the HUB&SPOKE topology as we currently have two HUBS in two DC's. 

The localized policy that has been attached to the HUB device is attached here, The localized policy that has been attached to the SPOKE device is attached here. The used CLI addon on the hubs, to set the loopback MPLS interface as QoS hub is also attached here.

On spoke interfaces adaptive QoS is set with Shaping rates upstream, the available upstream and downstream BW is set also. On the other side on HUB, we do not have adaptive qos turned on, only shaping on two interfaces, the one to Internet and the other one to out L3 switch (where all the MPLS services are comming together to the router). Attached them here also.

I hope I've provided all the info to get a clearer picture of the configuration. Thanks

Amar_Tufo
Level 1
Level 1

For future researches. We managed to resolve our issues, with the belowed mentioned:

1. Localized policy for both hub and spoke, they are different of course depending on the underlay.

2. The Localized policy on the spoke contained a QoS Forwarding class with 4 queue's, and also a ACL.

3. The Localized policy on the hub side contained the same ACl but a bit different, I'll explain below. The QoS Forwarding class was implemented through a CLI addon, was easier for us to bind it to the loopback interface.

4. shaping is applied on interfaces on hub and spoke, mpls and internet ofcourse.

The ACL's contains match conditions as:

1. protocol (6 for TPC,  17 for UDP)

2. port numbers

and action is accept with counter name and most important Class name which is configured in the QoS map, so to force the specific traffic into the proper Queue. ACL on hub are applied on the GiEth interface from the L3 switch where our MPLS traffic terminates, and on the spoke it's applied on VLAN1 as its in the VPN to the hub spoke.

The ACL on spoke defines source ports, and the acl on the hub side defines destination ports, depending on the traffic. 

With that we realized that the defined important traffic has bandwidth guarantied, and gets even more bw if other queues are not in use. 

Hopefully this approach will be usefull to someone one day.