cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
653
Views
0
Helpful
0
Replies

Unpredictable unequal mpls traffic engineering load sharing on ASR9000

yevgenyl
Level 1
Level 1

Hello,

 

In my company we are implementing a relatively simple mpls traffic engineering structure.
I will start by describing the basic setup.

 

Most of the TE tunnels are initiated from our core routers in our main pop to our regional pops through various paths. Please see the attached diagram.

 

For the implementation we run isis, ldp, rsvp, TE.
On ASR9000 core routers, under the rsvp configuration section there are the outgoing/incoming interfaces with the maximum bandwidth setting possible for the interface. In addition, under the "mpls traffic-eng" section "load-share unequal" is configured.

 

The TE themselves use explicit paths.

The usual setup is making two tunnels from each core router to each PE.

This is to ensure that no matter to what PE the traffic needs to go, or from what Core it comes, the SDH will always be properly utilized (Please see the attached diagram.):

 

core-A

-----

TT1: core-A -> PE-A

TT2: core-A -> core-B -> PE-B -> core-A

TT3: core-A -> PE-A -> PE-B

TT4: core-A -> core-B -> PE-A -> PE-B

 

And the corresponding tunnels on the second core router:

 

core-B

-----

TT1: core-B -> PE-B

TT2: core-B -> core-A -> PE-A -> core-B

TT3: core-B -> PE-B -> PE-A

TT4: core-B -> core-A -> PE-B -> PE-A

 

Now comes what I dont understand.

The ASR9000 core routers dont always distribute the traffic according to the tunnels load-share values.

It can be that one SDH path is congested, while others have 20% spare capacity.

Or the SDH paths would carry the same amount of traffic, even if on the TT which pass on specific SDH paths we configured the corresponding load-share values.
For instance, all tunnels which pass between core-A to PE-A have load-share of 750, and all tunnels which pass from core-A to Pe-B have load-share 0f 600. In reality, for some reason the distribution would be equal.

 

In most cases the load-sharing is "ok", but not perfect.

At other cases, it seems like not working, or just is not showing properly until a congestion of one the paths and then more traffic gets shifted onto the free paths (but not always). In all cases, we under utilize some paths and over subscribe others when we are at total utilization of 85% and above.

There are many thousands sessions going to the pops so i'm not worried about a few heavy sessions being stuck on specific path.

 

Is there a way to improve it?

I'm shy of dynamic paths as I cant guarantee that our vast network has correct isis metrics and other means to prevent tunnels taking the wrong path. Auto-BW requires dynamic paths, correct?
I'm also not sure that dynamic paths is the answer.

What can you suggest?

Thanks a lot

 

This is the TT typical syntax:

interface tunnel-te22221
 description TE_TUNNEL_TO_PE-A
 ipv4 unnumbered Loopback0
 load-interval 30
 signalled-bandwidth 0
 load-share 600
 autoroute announce
 !
 destination 2.2.2.2
 path-option 1 explicit name PE-A_via_CORE-A_PE-A

 

0 Replies 0
Review Cisco Networking for a $25 gift card