Posted on June 4, 2015 11:00 am
 |  Asked by Lukas Hubschmid
 |  1901 views
RESOLVED
0
0
Print Friendly, PDF & Email

Hello everybody,

I have configured a leaf-spine architecture (4 leafs, 2 spines), see attached image. L3 routing is enabled on leafs and spines and eBGP is used between them. To enable VXLAN (from VMware Hosts), multicast needs to be configured. So far, ip multicast-routing and ip pim sparse-mode is set.

Where do I configure the rendezvous point (RP) and which method would I use?
I would prefer to place it on the spine switches using anycast-rp, but one spine switches does not reach the other spine switch (as the route to spine1 is not announced to spine2 by the leafs through BGP).

Many thanks in advance!

Kind regards,
Lukas

Attachments:
Marked as spam
2
Posted by Cox Greg
Answered on June 5, 2015 5:07 am

I’ve gone thru trying to decide that myself recently, and noticed a few unexpected things.

Multicast Vxlan distribution is only supported on previous generation hardware, like 7150, it is not supported on newer chipset like 7050SX(without software updates that are not scheduled as far as I can find), only headend replication.

I would consider this more of a bug, but it seems the RP cannot be on a node one hop upstream from your Spine when your Spine is a VTEP(so your ”Shared Tree” cannot be one hop up from the VTEP with your ”Source Tree” still being down to your VTEP).

Personally I think the RP being 1 hop upstream from your Spines actually simplifies some things, when doing anycast(msdp).  From reading it seems there are caveats related to running anycast on Arista with synchronous paths between your msdp peers(potentially this is only on certain models, but just trying to remember).

Also when running headend replication, and your leaves and spines are vtep’s, joins that are within your vni’s do not get decapsulated and forwarded north of your spine.  It’s as if they have to traverse to a second hop like they have to perform a PenUltimateHopPop like function or else they  get dropped.

All of the above is from my testing on 4.14.7m on 7150S’s.  So I hope I am wrong, it’s been fixed on more recent versions, or there are more fixes scheduled.

Not sure if it helps, but hope it saves someone some time.

Thanks

Greg

Post your Answer

You must be logged in to post an answer.