1

We will be deploying into an environment intended for maximum resilience/redundancy and the capability for rapid scalability.

Context

This environment is:

  • 1 x NAS server with FreeNAS RAID
  • 2 x application host servers
  • 1 x managed layer 2 switch
  • 1 x firewall/router with 250Mb/s max WAN throughput

Both the NAS and the app servers are Citrix XenServer hosts, with the NAS holding the storage repositories and the app servers hosting the VMs. One of the main intentions for this is that, as we scale, we can introduce new hardware and repair or remove old hardware without interrupting the service by live-migrating between the app hosts (or the NAS for that matter, once we purchase additional units).

Question

Having not performed live-migrations in a production environment before, my concern is for the layer 2 switch. Will the switch be able to handle the fact that some mac addresses are being routed through port x one moment and port y the next?

For reference, the specific switch is here: http://uk.tp-link.com/products/details/cat-39_TL-SG3424.html

  • 1
    Generally the hypervisor will send a gratuitous ARP at the right time, so all you should need to do is make sure port security is off. I'm not familiar with this particular switch, though, so I'll leave it at that for now. – Michael Hampton Jan 17 '16 at 4:54
  • @MichaelHampton Thank you, I'll wait to see if anyone has something to add but that sounds like I won't have too much a problem. – Marvin Jan 17 '16 at 4:58
0

It will work well, live migration is standard in production environment. But your infrastructure is not so reliable as you think it is, there are many SPOFs in that. What will happen when your switch will die? Or even on port on that switch? What will happen when your NAS will die? If you need high availabiliy, every component of your infrastructure must be mirrored! You will need two mirrored NASes, two switches (and your servers should be connected to both of them!), two firewalles and two separated wan connection. There is no HA if you don't have it.

  • Thanks, seems like its no worries. We are aware that we do not have redundancy in our current environment; our budget is limited and I don't think we can justify the additional £2000-£2500 expense for the initial launch. After launch our first priority for purchases will be to achieve full redundancy with an additional switch, NAS, and then router. – Marvin Jan 17 '16 at 9:12
1

In addition to Ondra Sniper Flidr and the comment by Michael Hampton, I wanted to provide the official statement by Citrix. Quoting this pdf (didn't found this in HTML), chapter "XenMotion", page sixteen:

But in this example setup, the external real switch device is expecting the MAC address of the VM to be on one port, while it‟s actually just migrated to another port. In this case, the last step of the migration is for the destination host to update any external devices with a gratuitous ARP packet. A “garp”, is nothing more than a regular ARP request, but with the MAC and IP address already filled-in. This serves as an update to any external devices‟ arp cache, and there would be no response.

  • Thanks, I had missed this. It's always nice to see official documentation. – Marvin Jan 17 '16 at 10:16

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.