FlexFabric question regarding redundancy

[chuckk281]

Predrag wanted some redundancy clarification:

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

I need help to clear in my head once for all time about redundancy in FlexFabric environment.

I  have bay 1&2 FF and both have two uplinks for LAN and two for SAN.

Now, when we talk about LAN, since I don’t have too many uplink ports I should use VLAN mappings, so, I need SUS, right.

Now, if I want redundancy, do I have to create two SUS sets, for example SUS named Trunk_1(with both ports inside) for 3 associeted networks (VLAN 101,102,103) and also second SUS Trunk_2 for other three vNETs (201,202,203)

Is this right example if I want to get redundancy if some serevr port or uplink fails? Or I didn’t understoood the VC purpose in right way?

And if I am right how these VLANs should be defined on external switches, so I can in any time have connectino on left and right side of FF?

Because when I look at the guides there are many different examples but mostly they have for example Prod_1 and 2, Backup_1.

I know SAN connection, and know that ports 1 and 2 b will be FcoE, but LAN is little bit confusing with all those options.

My main question is how to get redundancy and what is best practice?

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

From Robert:

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

You are correct about the need for a SUS and most likely Vlan mapping.  There is concern about the six Vlans you listed.  If the Vlans 101, 102, 103 are connected to one switch, and Vlans 201, 202, 203 are connected to another switch, then you will not be able to have redundancy using only two LAN uplink ports from VC. 

If all six Vlans are connected to the same switch or switch stack, then you have two choices for redundancy.  The difference in the two choices will determine where the failover occurs and total aggregated VC bandwidth. 

With an active/active configuration, the failover will be managed by the OS teaming or bonding.  And as the name implies, packets will be forwarding on the uplinks for both SUS or VC network.  SmartLink will need to be enabled on all VC networks within each of the SUS.  SmartLink will give VC the ability to disable downlink ports which are associated with failed uplinks for a SUS or VC network.  The teaming/bonding driver will detect the unlinked port and failover to the other team/bind member. 

With an active/standby configuration, VC manages the failover.  And again as the name implies, only uplinks for one of the SUS will be forwarding packets at any given time.  If all uplinks for a SUS or VC network fail, then VC will re-elect a new active.  I use the term re-elect because there may be multiple remaining “good” paths to choose from.  The internal VC cross links will be used to provide the path for traffic originating on blade ethernet ports that are connected to the VC module whose uplink paths are in a ‘linked/standby’ or ‘unlinked’ state.

Below is an example of an active/active configuration where there are two SUS.  All vlans from a single switch are defined on both. 

Trunk_1 (enc0:Bay1:X1):

Vlan 101

Vlan 102

Vlan 103

Vlan 201

Vlan 202

Vlan 203

Trunk_2 (enc0:Bay2:X1):

Vlan 101

Vlan 102

Vlan 103

Vlan 201

Vlan 202

Vlan 203

Below is an example of an active/standby configuration:

Trunk (enc0:Bay1:X1, and enc0:Bay2:X1):

Vlan 101

Vlan 102

Vlan 103

Vlan 201

Vlan 202

Vlan 203

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

And input from Dan:

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

Active/Active (2 SUS) vs Active/Passive (1 SUS) is not really a redundancy choice.

The reason you would chose one or the other has to do with the amount of Bandwidth you need leaving the chassis and whether the traffic overall is primarily East/West or North/South.

Both designs can be made fully redundant.

Active/Active you just enable SmartLink on the networks (VLANs) and as mentioned use the Server side Teaming.

Active/Passive you just add uplinks from both VC1 and VC2 and then VC will handle the failover.  You should still use Server side teaming in case of a VC failure (as opposed to something upstream failing)

Further Explanation of East/West vs North/South

Definitions:

               East/West means the traffic is going from server to server within the same VLAN (Network). (A Different VLAN would require the traffic to leave the chassis for Layer 3 routing)

               North/South means the traffic is going from a blade server to some other device outside the Chassis/Stack.  Anything on a different VLAN would be considered North/South as mentioned before because of Layer 3 routing.

If your traffic is primarily North/South, I would recommend 2 SUS (1 per VC) as you double the amount of ACTIVE bandwidth going between the chassis/stack and the rest of the network.  Its up to the Server OS/Hypervisor to be able to make full use of this bandwidth.  VMware does this fairly well.

If your traffic is primarily East/West, then Server Network Teaming mixed with 2 SUS can cause your traffic to become North/South.  This is because INSIDE VC the 2 VLANs (101A and 101B) are seen as different networks.  If Server 1 sends a packet on 101A and the destination MAC Address is registered on 101B, then the packet will have to leave the chassis, go through an external switch and then come back into the other VC to be delivered.

If you had used a single SUS for this traffic, then the internal Cross Link is used to allow a server transmitting on vNIC 1a to talk to another server on 1b.  With 2 SUS the Crosslink is barely used.

Now the other option is a mix as seen in our vSphere Design Reference.

You use 2 SUS for external communications and then use a single Network for things like vMotion and FT traffic.  This doubles the uplink bandwidth for North/South while keeping the heavy hitters on your East/West traffic contained within the Chassis/Stack.

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

Any other comments on the whole Active/Active vs. Active/Passive redundancy question?