Trunking across redundant switches 1

[jbarelds]

Hi all,

I'm in the process of writing a blueprint for a P2V (Physical-to-Virtual) migration for the majority of our servers (about 82). VMware ESXi is to be used as a virtual platform, the VM's will be hosted on 2+2 ESX servers (2 server rooms).

I'd like some help designing the physical network infrastructure. In ESX it's possible to create teams of physical nics. Lets say I want to team 4 physical nics on an ESX server, to allow a maximum of 4Gbit throughput. I want the switches to be redundant, so that if one switch fails, there's still 2Gbit available on the other:

As I'm not a Cisco guru, I'd like to get some feedback if this is possible.

 

[VinceWhirlwind]

Your "trunk" is 802.3ad/etherchannel, so the other end of those 4 ports also need to be aggregated in the same way.
That means Switch1 & Switch2 will have to be members of a virtual switch/switch stack, which is exactly how I've seen these ESX aggregated links used.

Your alternative is to look at the capabilities of your virtual switch within the ESX server and see if you can use spanning tree or some other kind of ugliness.

 

[unclerico]

IMHO go with Cisco 3750's and a Redundant Power Supply (RPS 2300) for this. From the 3750 product literature:

Cisco StackWise technology is a revolutionary stacking architecture that brings high levels of resiliency, automation, and performance to stackable switches. With Cisco StackWise technology, customers can create a single, 32-Gbps switching unit with up to nine Cisco Catalyst 3750 Series switches.

Using 3750's will allow you to do as Vince says and that is to make a stack of switches look as one switch and let you create a 4-port EtherChannel (or more depending on your needs) across the members of the stack. You also manage all switches in the stack as one switch. This will give you that resiliency that you are looking for plus ease of management.

One thing to watch out for is that in ESX you need to make sure your virutal switch is in agreement with the physical switch that it is connected to in how load sharing will work. For example, if your vSwitch is set to use IP Hash (vSwitch Properties -> Load Balancing) then you need to make sure that the physical switch is set to use IP Hashing also otherwise you'll get connectivity issues. Using a Cisco switch as an example you'd enter the following command:

Switch(config)# port-channel load-balance src-dst-ip

I'm also curious as to what the specs are for the hardware you're going to be using.

I hate all Uppercase... I don't want my groups to seem angry at me all the time! =)
- ColdFlame (vbscript forum)

 

[jbarelds]

Vince, unclerico,

Thanks a lot for the replies above. All I needed to finish my blueprint. And thanks for the tip about the load sharing method, will keep that in mind.

unclerico, about the specs of the hardware. In my blueprint I refer to a Dell PowerEdge R805 equipped with two quad-port intel Gbit cards, two single channel 4Gbit FC HBA's, 64GB memory and two quadcore opterons.

 

[unclerico]

Looks like you've got a nice setup there. Of course you're going to crash and burn by using Dell ) ...just messin with ya of course. Good luck with your project.

I hate all Uppercase... I don't want my groups to seem angry at me all the time! =)
- ColdFlame (vbscript forum)

 

[burtsbees]

If you have two 4-port GB NICs in the server, why not make two redundant 4GB pipes to each switch, so that you will have 4GB available no matter what in case of failure?

Burt

 

[jbarelds]

Hi burtsbees,

VMware Management and Vmotion traffic both need a dedicated interface (not mandatory, but highly advisable). Leaves me with 2 spare interfaces, not a whole lot considering future growth and the addition of VMware FT in VI4 later next year.

 

[burtsbees]

I see.

Burt