The Cloud and High Availability

One of my new favorite podcasts is RunAsRadio with Richard Campbell. The June 14, 2023 edition was "High Availability in 2023 with Allan Hirt".

You can see Allan Hirt's credentials on LinkedIn.

Here's the synopsis of the podcast:

What does high availability look like in 2023? Richard chats with Allan Hirt about his work with high-availability solutions today - not just on-premises but also in the cloud. Allan talks about the frustration folks had with moving workloads in the cloud during the pandemic panic, lift-and-shifting workloads focusing on getting things working quickly rather than cost-effectively. The results can be costly, to the point where some folks considering moving back off the cloud again - but does that make sense? Allan talks about creating high availability efficiently wherever you want to run your workloads!

Richard and Allan covered availability as manifested in the cloud. Allan is a SQL Server guru who works for Pure Storage.

Here are some excerpts.

7:31 But when it comes to high availability, really, it's tough.

And cloud isn't a silver bullet for availability.

7:58 Like everybody assumes that there's this big red button that says I've got high availability in any cloud. We're not just talking Azure, we're talking AWS and GCP.

And the cloud doesn't eliminate traditional availability issues.

8:54 You know, like your platform layer, meaning Azure or GCP, like I can spin up this VM in another region when something happens is great. But the problem then winds up being it doesn't account for things like transactions in a SQL Server installation. So all the traditional things we worry about on premises still exist up in the cloud with IaaS. It's PaaS where things get different.

And the cloud introduces new issues.

13:15 … but now we need to think about placement of our VMs. We need to think about like ensuring that the VMs don't sit on the same host or in the same rack, which means setting certain things differently in your cloud providers.

And old issues don't go away.

15:10 … I appreciate the idea that if you're responsible high availability, … that doesn't go away with the cloud. … You still are responsible for [high availability] and you have to work through the additional complexity.

And monitoring gets even more important.

23:11 And you really don't have a view of is it functioning well. And then you need somehow you need to be able to monitor the stuff remotely and realize we have no insight into what's going on into our systems. I saw an uptick in implementation of like a lot of the third party monitoring tools or SCOM … because people "Like we need this stuff." Whereas they didn't put too much importance on that.

But cloud providers likely provide higher availability than you could.

28:15 I still think the cloud is up more than any data center I ever ran was. What I've actually told customers is Microsoft, Amazon, and Google can outspend you in building a data center. And a 24 hour staffed NOC is normal for them and expensive for anybody else.

But just moving to the cloud doesn't guarantee more benefits.

30:46 So I can take my on premises VMware infrastructure, vMotion it up to Azure, AWS, GCP and vice versa all day long, right. … I think for some folks that's a good solution. … It gets you to the cloud. It works well sort of. … And so it's always interesting to see things like this is where it's more of a on premises play in the cloud.

If you're not good at on premises, you won't be any better on the cloud.

34:56 If you sucked at virtualization on premises, you're generally going to suck at stuff in the cloud. Not always, but it's generally true because if you couldn't size and you couldn't deploy well on premises, how are you going to automate and do all this cool dev ops, dev stack ops stuff. If you had no skills to do it on premises, it's not magically going to happen in the cloud.

AWS - Amazon Web Services

GCP - Google Cloud Platform

IaaS - Infrastructure as a Service

PaaS - Platform as a Service

SCOM - Microsoft System Center Operations Manager

NOC - Network Operations Center