I have a consulting client with a relatively small vSphere implementation. Comprised of just four ESX servers, this client typifies the small implementation.
Yet while small, they’ve been smart in how they’ve constructed that virtual infrastructure. Their ESX servers connect to a central SAN, enjoying all the VMotion benefits that working off a shared Datastore brings.
I was taking a look at their configuration a few weeks ago when I noticed that their configuration didn’t include creating a cluster. Each of their hosts was displayed individually in the vSphere Client, showing their virtual machines by host rather than as part of a larger resource pool.
I asked this client, “Wouldn’t you want to create a cluster out of these hosts, so that you can share the resources between them?”
Their response was one I wasn’t expecting, “Well, we thought about that. But we found that we really don’t use VMotion pretty much ever. We know that we can use VMotion, and sometimes we do. But, our performance is acceptable so we don’t need DRS, and we find that that we’re really never doing activities that require us to relocate or re-balance our virtual machines.”
That response really got me thinking about the uses of VMotion in contrast with all the publicity its feature gets in the IT press:
- You use VMotion prior to rebooting a host.
- You use VMotion to re-balance load (often through DRS’s automated processes)
- You use HA after an unexpected failure.
Each of these three use cases are important, but at the same time they don’t occur all that often. ESX’s Linux roots mean that host reboots are a rarity. Loads indeed need to be re-balanced, but this client (and others which I asked after having this conversation) doesn’t see the need to add its extra complexity. HA indeed protects hosts after a failure, but other than the occasional full power outage hosts don’t tend to fail too often.
This brings me to this post’s question: When was the last time you VMotion-ed? Was it recently? Do you do it often? Or, now that your virtual infrastructure is up and running in steady-state, do you find that you really don’t use it all that often? Give me the low-down in the comment field below.