Behaviour AAR with sla

[dijix1990]

I know that sd-wan edge can change output interface based predefined sla, it's perfect, I use it. Now I'm interested for example

I have some cEdge's. Every has two colors

• biz-internet
• public-internet

Of course every cEdge has different latency some has 30ms for both isp's and some 70ms for both isp's

Can sd-wan change prefered color because of latency? for example, 

I have AAR policy load-balance with common sla:

• Latency - 100ms
• Loss - 5%
• Jitter - 30ms

and something happened with biz-internet for cEdge-1

cEdge-1 now has:

• for biz-internet Latency=60ms (was 30ms) / Loss=0% / Jitter=5ms
• for public-internet  Latency=30ms (was 30ms) / Loss=0% / Jitter=5ms

and I wouldn't want to cEdge-1 sent traffic via biz-internet because of Latency=60ms. What can I use for implement this behaviuour? Only use different AAR policy for every sites?

[balaji.bandi]

So are you looking to do this all routers or only some sites ?

you can set the latency  above 50ms so that will be less prefered for traffic. 30ms to be prefered path.

check below guide :

https://www.cisco.com/c/en/us/td/docs/routers/sdwan/configuration/policies/ios-xe-17/policies-book-xe/application-aware-routing.html

Note : if you looking only site by site, then you need to change the template in specific to site. if you using command template for all device.

[dijix1990]

For all sites, now I have about 50 sites with different latency and if I decrease sla latency for all sites it means that for some sites sla for aar won't be meet and for this sites will work data policy. If aar policy will see varience only for recently changing latency or loss it will be good, but as I see it's impossible right now

My topology is hub and spokes (two different dc's) 

[balaji.bandi]

if each site have different requirement, then you need different templates.

we have some application running over 200MS and works fine, the only question here the latency matter is Voice traffic.

[Kanan Huseynli · Accepted solution]

Hi,

normally you should then use different AAR policy.

Here one interesting point exists. If latency 100ms is OK for application, you should (normally) not care it is 30ms or 60ms. Your path still satisfies SLA, no? I understand you want over super bestpath (lower latency in this case), but there is no such logic.

You may decrease SLA and if for some sites none of SLA is met, then you may have "fallback-to-the best" logic with latency variance to avoid dampening. Or backup-preferred-color (which works only when no path meets SLA) you can choose based on "statistical analyses". But no deterministic result is guaranteed, of course.

Personally, for me if path meets SLA then it should be OK even though there is "ideal" - better path. Applications will work without any issue (otherwise what is the meaning "meeting SLA"?)

[dijix1990]

Yes, it's true if path meets SLA then it should be OK despite sla latency is 60ms or 80ms. I'm thinking about it because when remote branches works via horizon or wyse clients and suddenly latency for one channel increase about 40ms from 30ms It's noticeable for clients, they start to complain about slowly response. 

[Kanan Huseynli · Accepted solution]

Then, better to have site specific AAR policy with more strict values (like 50ms SLA), if path latency values jumps to higher for a long time (of course I mean based on AAR poll interval etc.) then better path will be used