esx redundant switches question

[Hi5]

Hi Guys,

Please help me to clarify this.

I have attached simple diagram showing configuration.

Question: Do I need to connect two switches together in order to get VM1 to talk to VM2 and vice versa and provide switch redundancy?

Thanks!

[logiboy123]

The simple answer for your basic diagram is yes.

The complex answer is dependant on the equipment you are using, the configuration of your hosts and how you want your virtual machines to communicate.

What exactly are you trying to achieve?

[Hi5]

Thanks for quick reply!

I want to have redundant network for VMs without interconnecting two physical switches.

Let’s say we have 2 ESX hosts with 40 VMs on each.

Each host has the following config:

All VMs are connected to a single vSwitch

vSwitch has multiple portgroups with different VLANs for VM network, VMotion and service console

2 physical NICs are assigned to the vSwitch. They are NIC Teamed (Route based on the original port ID)

One pNIC1 goes to pSwitch1 and pNIC2 goes to pSwitch2

No extra configuration on physical Switch except VLAN tagging

I still don’t quite understand why switches have to be connected to each other in order to provide connectivity between VMs.

Let’s say that switches are not connected to each other:

If everything is up and running in my diagram then there 2 paths for each VM to communicate to each other.

If any single physical switch/link/NIC fails there is still one path exists between VM1 and VM2

Thanks!

[bulletprooffool]

I like the idea and the design, but if you have configured load balanced vSwitches and the vSwitch on Host 1 decides it wants to load balance via the port  attahced to pNic1 - while the second host decides to use pNic2 (e.g. there is a fault on pNic 1 etc) you are going to run into pronblems.

Imagine a scenario where host 1 has a failure on pNic1 and host 2 has a failure on pNic 2 - you now have no redundant route and traffic can not route between VMs on these vSwitches.

[Hi5]

So..

vSwitch chooses one pNIC to send/recieve VM traffic and if it cannot reach VM on another host it does not try to send traffic through another available pNIC

Is this correct for my diagram and design?

Thanks!

[Josh26]

Hi,

Yes I believe that for your diagram and design, there is no redundancy.

Imagine host2 loses the connection to switch1.

If host1 sends any traffic via switch1, it can never land on host two. You bought two switches for redundancy didn't you?

A server can detect when a directly connected path has failed. For example, in the above situation, host2 will send everything to switch2 which it knows is online. But a server cannot know about "upstream" failures.

[Hi5]

No I haven't bought them yet that is why I am asking :smileyhappy:.

Thanks guys for your help. It is now well clear to me that switches have to be interconnected for connectivity and redundancy.

[Josh26]

If you would allow me to complicate matters further - the best practise would be to join those switches twice via etherchannel, for even better redundancy - and 2Gb throughput should it be required.