Windows Azure In-place Upgrades (Year of Azure – Week16)

On Wednesday, Windows Azure unveiled yet another substantial service improvement, enhancements to in-place upgrades. Before I dive into these enhancements and why they’re important, I want to talk first about where we came from.

PS – I say “in-place upgrade” because the button on the windows azure portal is labeled “upgrade”. But the latest blog post calls this an “update”. As far as I’m concerned, these are synonymous.

Inside Windows Azure

If you haven’t already, I encourage you to set aside an hour, turn off your phone, email, and yes even twitter so you can watch Mark Russinovich’s “Inside Windows Azure” presentation. Mark does an excellent job of explaining that within the Windows Azure datacenter, we have multiple clusters. When you select an affinity group, this tells the Azure Fabric Controller to try and put all resources aligned with that affinity group into the same cluster. Within a cluster, you have multiple server racks, each with multiple servers, each with in turn multiple cores.

Now these resources are divided up essentially into slots, with each slot being the space necessary for a small size Windows Azure Instance (1 1.68ghz core, and 1.75gb of RAM). When you deploy your service, the Azure Fabric will allocate these slots (1 for a small, 2 for a medium, etc…) and provision a guest virtual machine that allocates those resources. It also sets up the VHD that will be mounted into that VHD for any local storage you’ve requested, and configure firewall and load balancers for any endpoints you’ve defined.

These parameters, the instance size, endpoints, local storage… are what I’ve taken to calling the Windows Azure service signature.

Now if this signature wasn’t changing, you had the option of deploying new bits to your cloud service using the “upgrade” option. This allowed you to take advantage of the upgrade domains to do a rolling update and deploy functional changes to your service. The advantage of the in-place upgrade, was that you didn’t “restart the clock” on your hosts costs (the hourly billing for Azure works like cell phone minutes), and it was also faster since the provisioning of resources was a bit more streamlined. I’ve seen a single developer deploying a simple service eat through a couple hundred compute hours in a day just by deleting and redeploying. So this was an important feature to take advantage of whenever possible.

If we needed to change this service signature, we were forced to either stop/delete/redeploy our services or deploy to another slot (staging or a separate service) and perform either a VIP or DNS swap. With this update, much of these imitations have been removed. This was because in the case of a change in size, you may have to move the instance to a new set of “slots” to get the resources you wanted. For the firewall/load balancer changes, I’m not quite sure what the limitation was. But this was life as we’ve known it in Azure for last (dang, has if really been this long?)… 2+ years now.

What’s new?

With the new enhancements we can basically forget about the service signature. The gloves are officially off! We will need to the 1.5 SDK to take advantage of changes to size, local storage, or endpoints, but that’s a small price to pay. Especially since the management API also already supports these changes.

The downside, is that the Visual Studio tools do not currently take advantage of this feature. However, with Scott “the Gu” Guthrie at the helm of the Azure tools, I expect this won’t be the case for long.

I’d dive more into exactly how to use this new feature, but honestly the team blog has done a great job and and I can’t see myself wanting to add anything (aside from the backstory I already have). So that’s all for this week.

Until next time!