TOPIC:

Context

We are using Dell PowerEdge R410 servers. These servers have two 1Gb/s network ports on the motherboard and one PCIe slot.

The PCIe slot is in use by the RAID controller so we have no means of providing additional network ports other than something like a USB NIC (note: if we did use a USB network card then it would be USB 2.0 only which means that we would not get Gigabit speeds). Whether any of the USB to ethernet adapters would play nicely with a Dell PowerEdge with XenServer hypervisor... I have no idea.

Currently I have setup the ports thus:

• port 0 is connected to the network and is transporting all management + LAN + WAN traffic
• port 1 is connected to the paired high-availability server and is handling comm traffic between them

This obviously has two major problems:

• No cable/port redundancy - a faulty cable or network interface will completely sever the connection to one or other network.
• No separation between management, LAN, and public traffic

Quesion

What advice do you have about the best network setup with the limitations described above?

As I see it, these are the options:

1. Keep it as it stands
Port 0: management + LAN + WAN
Port 1: HA comms

2. Try to use USB adapters for redundancy
Port 0: management + LAN + WAN
Port 1: HA comms
Port 2: management + LAN + WAN (USB)
Port 3: HA Comms (USB)

3. Try to use USB adapters for the HA comms
Port 0: Management + LAN
Port 1: WAN
Port 2: HA comms (USB)
Port 3: HA comms (USB)


Bonus Question

If the network connections remain active but the paired high-availability servers lose their storage connection, will there be any 'split-brain' related problems?

At first I thought there would be but then I thought that the two servers would still know which VMs are running on the other server, so they wouldn't try to run the same VMs. In which case there shouldn't be any data conflict so, when they are reconnected, the data should sync correctly. But then I might be missing the point entirely.


Thank you.

Thanks for your post. You don't really have many options here.

In response to your qustions:

1. Keep it as it stands
I think that is your best option, although not ideal

2. Try to use USB adapters for redundancy
I have no experience with USB Ethernet adapters and would perceive them to be rather slow. I would not recommend USB for the data replication link for 2 reasons: speed AND potentially fragile compared native Ethernet. They would probably be OK for VM traffic.

3. Try to use USB adapters for the HA comms
I would steer clear of using USB for replication.

Regarding your bonus question, if you lose your replication interface you will not have any split brain issues nor will your data be affected - however, the slave server uses the communications link to access disk images for any VMs running on the slave. So, in this scenario, losing your replication link equates to your slaves losing access to their disks.