BladeSystem Virtual Connect
cancel
Showing results for 
Search instead for 
Did you mean: 

Supported Virtual Connet (VC) Active/Active Configuration

chuckk281
Trusted Contributor

Supported Virtual Connet (VC) Active/Active Configuration

Zubair had a customer VC configuration to support an Active/Active setup:

 

****************

 

One of my customer having C7000 enclosure equipped with 2 x Flex 10/10D modules. From each Flex 10/10D module one uplink (configured as a SUS, currently having only one uplink cable) is created and link to two different Juniper EX4550 switches.

Both links are active as a separate SUS. At server end NIC teaming will be used (some NIC teaming will be configured Active/Passive and others will be as Active/Active). Below mentioned is the implemented scenario. Is this a supported configuration for VC ?

 

Customer’s network team has raised their concern about creation of network loop while  connecting two VC modules in active/active scenario to Juniper switches, IS this a supported configuration and what configuration should be done on Juniper switches ?

 

webpic374.png

 

***************

 

Reply from David:

VC modules are layer 2 and do not participate in Spanning Tree.

It would not create a loop with bonding/teaming.

The only way that I have heard of loops in the past was with NIC bridging.

 

We now have in the latest codes NIC loop protections that will shutdown a downlink(to NIC) if VC detects any loops.

 

And from Brian:

Remember since you are doing the dual SUS route the lower level vnets associated with those sus are separate broadcast domains.

 

For example

vNET A(vlan100) -> SUS-A

vNET B(vlan100) -> SUS-B

 

The network guys are assuming the vNET A and vNET B  vnets are the same vlan inside of Virtual Connect.  However they are not.  When traffic leaves those vlans and go up to their SUS they will get a VLAN 100 tag but internally they are different entities.  In fact to communicate between VNET A and VNET B a packet must leave the VC module and go up and then come back down (hairpin turn)

 

My rule of thumb is that if your environment is mostly server to server traffic in the chassis\domain then an active\passive sus design is better because you will keep more traffic in the chassis. (Plus it’s simpler to implement)  If your traffic is mostly server to the rest of the datacenter than an active\active design is better as it improves overall chassis throughput. With the insane amount of bandwidth available now I am seeing more and more customers go the single SUS route…

 

************

 

Comments?