cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
4880
Views
10
Helpful
6
Replies
Highlighted
Beginner

Multicast in Nexus vPC environment problem

We have a network consisting of a 2x 5k core, 2x 5k DC distributions, 2x2 5k DC L2 aggregation and 2x 3850-48XS Campus L3 distribution (see attachment). The rendezvous points (RPs) are being shared on the cores switches via MSDP and all L3 devices point to the shared anycast-rp IP address as to their single RP IP address with an ACL limiting the stream to the 239.0.0.0/8 range.

 

The core switches talk to every L3 distribution switch exclusively via L3 routing (EIGRP). The DC distributions switches do create vPC connections to the end-of-rack 5k switches which connect several 2k FEXes. The multicast source is connected single-homed to one of the FEXs and the receiver on the Campus distribution. The source is streaming 10 different streams mapped to 10 different mcast addresses (all in 239.0.0.0/8 range). 

 

The problem is, that I can randomly watch just every other multicast stream. In other words my VLC app is able to open/join just 4 to 5 streams of the streamed 10!

 

I have troubleshooted very deeply and all seemed pretty well. Than I got a thought to disable one L3 device after the other, step-by-step, since this behavior felt like a redundancy problem. After I shut down the secondary DC distribution switch (DC2-DC_distri) - BINGO! >>> every stream was funcitoning immediately properly and the switching between programs went blazing quick.

 

Do you have any thoughts, what could cause such abnormal behavior? After re-enabling the DC2 distribution, the problem came again! So just a plain reboot of a networking device didn't solve the problem. Just the shut down of this particular device solved it..

 

The configuration is prety simple. Every L3 interface has "ip pim sparse-mode", EIGRP routes everything to everything - basically every destination has two equal paths to get to. A permanent ping from source to destination was working stable for hours without loosing a single echo. So it has to have something to do with multicast routing or some strange vPC limitation. 

 

Thanx for every single hint! 

 

 

 

1 ACCEPTED SOLUTION

Accepted Solutions
Highlighted
Beginner

Probably - I FOUND THE SOLUTION!

I implemented following command on both L3 DC Distribution switches:

 

SWITCH(config)# vpc bind-vrf <TV_VRF> vlan <not_used/defined_VLAN>

 

DC1-DC_Distri(config)# vpc bind-vrf IPTV vlan 4000

DC2-DC_Distri(config)# vpc bind-vrf IPTV vlan 4000

 

Just implement this command on both L3 vPC Nexuses with the VRF where multicasts are being used and choose a VLAN which you HAVE NOT defined! It should work then.

 

 

View solution in original post

6 REPLIES 6
Highlighted
Contributor

This is a very common issue with layer 3 and VPC. You’d be better off with straight layer 3 interfaces and do ECMP. What happens has to do with hashing. Some packets go to the normal packet path via SW1 and others (because of hashing) will go via the other path via SW2. When the other switch see the packet it sends it over the peer-link which to SW1 which will not allow any packets down member ports. I’ve run into this issue several times.
Highlighted

Hey Rick. Thanx for the fast answer.

 

Couple of questions

1. What exactly do you mean by "straight layer 3 interfaces and do ECMP"? Every L3 distribution switch is connected via L3 P2P networks (all are /31 because of IP address saving) to the core. I am using

  • L3 port-channels with sub-interfaces (becauce of several VRFs) between DC distributions (Nexuses) and Cores (Nexuses)
  • L3 VLAN P2P (/31) transit networks between Campus distris (Catalysts) and Core switches.

 

2. I found here NX5600 PIM - Bind-VRF Configuration to vPCs a small chapter of configuring Bind-VRF which could possibly solve this issue. But the syntax is described pretty weird. I am not sure on which device(es) to configure it, which VRF and which VLAN to pick as the right one. Do you know this feature and/or do you have experience with it?

 

Thanx

Highlighted

What I meant to say, was the layer 3 interfaces will not participate in vpc and are regular L3 “no switchport” ports. With binding VRF’s I don’t have any experience with.
Highlighted

Do you mean Rick, to remove the vPC port-channels between aggregation and distribution in the DC and make routing between these two layers?

Highlighted
Beginner

Probably - I FOUND THE SOLUTION!

I implemented following command on both L3 DC Distribution switches:

 

SWITCH(config)# vpc bind-vrf <TV_VRF> vlan <not_used/defined_VLAN>

 

DC1-DC_Distri(config)# vpc bind-vrf IPTV vlan 4000

DC2-DC_Distri(config)# vpc bind-vrf IPTV vlan 4000

 

Just implement this command on both L3 vPC Nexuses with the VRF where multicasts are being used and choose a VLAN which you HAVE NOT defined! It should work then.

 

 

View solution in original post

Highlighted

Awesome, let me know how it works. Thanks for sharing.