My friend Brent Ozar recently posted that VMWare HA isn’t sufficient for SQL Server high availability.
I agree to a point. And at the same time, I think he’s taking his view to an extreme. That’s where this tweet came from.
If you’re running a system that requires the proverbial “five nines” then Brent is absolutely right. If you’re running a database server for a stock trading platform, a hospital, or some other system that manages people’s lives or the economy of a nation, you need more than VMWare high availability.
Here is where I start to disagree with Brent. How many of us run systems that actually require that level of uptime? Most of the systems that I’ve encountered in my career can tolerate a few minutes of downtime. From time to time things will go down. Yes, it will be painful, but it can be managed.
In my e-mail exchange with Brent, he made a really good point about patching. SQL 2012 has required quite a bit of patching as of late. In my environment, we have a planned, scheduled maintenance window every month. Our clients know that we will be taking systems down during that window. Still, Brent makes a really good point.
Historically, HA technologies have protected us from hardware failures and the occasional corruption of an operating system drive. With running systems on virtual hardware where all storage lives on a SAN, the risk of both of these concerns is mitigated a lot more. I honestly haven’t seen a VM get the dreaded BSOD in the past few years. Maybe I’m just lucky.
For many of us, we don’t need 99.999% uptime. And most of us can’t provide that on small company budgets. What VMWare HA does give us is just a little more availability without breaking the bank. It means that downtime is rare instead of unheard of. DBAs hate downtime. I get that. But at what point does this become an academic exercise? We need to align our uptime expectations with business requirements and technology costs.
My point is that we need to balance business need, budget, and reality. When the business says they need “five nines” they tend to scale back their requirements after seeing the costs.