Multiple default gateways on a single ESXi host?

[StageCoach20111]

If I am using virtual distributed switch, with NFS and management ports, can I create two separate default gateways, one for NFS and one for the management network?

[a_p_]

Please take a look at http://kb.vmware.com/kb/2001426 to see whether this helps.

André

[TheVMinator]

Does the same principle in this article apply to having NFS traffic on a separate subnet / vlan from management traffic?

[a_p_]

It actually applies if - for whatever reasons - you need more than one VMkernel port group (subnet) routed.

André

[StageCoach20111]

Great - it says you can do this using host profiles in 5.1 but it doesn't say how - any ideas on that?

[chriswahl]

A R wrote:

If I am using virtual distributed switch, with NFS and management ports, can I create two separate default gateways, one for NFS and one for the management network?

You are limited to one gateway.

NFS traffic should not be routed - put the vmkernel port IP on the same subnet as the NFS export's IP.

[StageCoach20111]

Using a non-routed subnet for NFS traffic is definitely best practice.  However, it seems from the article that VMware is saying that you can add a static route.  Notice, not a default gateway, but a static route at the command line.  If you are required to temporarily set up access to an NFS mount on a different subnet, it seems like the article is saying you should be able to do that with a static route.  Thoughts? 

[TheVMinator]

Whether or not it is NFS that is in question, the article does seem to indicate that multiple static routes are possible, so if you did want to use NFS routed or any other second vmkernel port routed, it seems that it should be possible, unless that article is incorrect.  At least that seems to be the purpose of the KB, to show that it is at least possible...but is it?

[rickardnobel]

Routed NFS are now supported, but with some heavy requirements, see this:

http://blogs.vmware.com/vsphere/2012/06/vsphere-50-u1-now-supports-routed-nfs-storage-access.html

[chriswahl]

They are still a very bad idea, often being a nightmare to troubleshoot and provide performance support. Why introduce multi-hop latency to your storage protocol?

[StageCoach20111]

There is no technical reason to do this and it is a bad design but sometimes if you are not the one making the decision you have to follow orders...

[rickardnobel]

Chris Wahl wrote:

They are still a very bad idea, often being a nightmare to troubleshoot and provide performance support.

I totally agree. It could however be situations where you might want to access some remote ISO store or something through NFS and it is interesting just to know it is possible.