How easy is it to get started with Windows Azure Infrastructure as a Service? If I am honest: Almost too easy.
It would be fair to say that Microsoft’s cloud platform has not enjoyed the same high profile as Amazon’s efforts, possibly because of the order in which the various services under the Azure banner have been presented doesn’t fit with the regular computer user’s conception of what a cloud might look like.
Azure has been much more of a developer’s resource right from the word go, emphasising the scale-out aspects of cloud utilisation with instances of compute resource you can latch on to and drive from Visual Studio, and a wide range of Microsoft’s other branded products living inside the Azure envelope – much more SaaS than anything else.
The biggest lack so far has been the disconnection between the consumer services that happen to use Azure as their back end (like SkyDrive) and the regular Joe doing things on their home PC. The daily life of a Microsoft person involves hundreds of interactions with cloud storage, document sharing and other Azure goodies every day, because Microsoft internal IT drinks its own brew, and that means that what, formerly, would take several careful evenings of research to get your machine Azure ready, is now delivered in a single login process.
If you are a keen Microsoft watcher then you will be looking for some mention of SystemCenter 2012. This is the heart of the company’s method for managing its own IT, and is an impressive achievement at a number of levels – but ease of adoption certainly isn’t one of them.
Talking to Microsoft seniors it’s very clear that they see SystemCenter as transformative, not just for IT types but for the businesses which depend on them ever more strongly. There is no chance that something this capable is also going to be a few clicks and a bit of downloading to actually implement: Possibly a large part of the slow adoption of Azure has been that people have until now, associated it with the intimidatingly complex bulk of SystemCenter 2012 and felt there was no quiet, experimental, low-risk way into this arena.
That’s by way of explaining why “Azure” may have been launched some time back, but only now is emerging in ways that should interest and indeed amaze the traditional IT infrastructure person.
IaaS on Azure has been around for a while too – I’ve had a server running just under the chargeable tier for a year or more, myself – but this new 2013 update marks a major innovation and a massive change in the way that Windows servers can be provisioned and configured. If you move fast and get to the free trial pages for Azure, you can look how long it will take you to put up a complete Virtual Machine with a globally-routable, usable domain name and an easily accessed RDP connection.
Compare this to the kind of build delays typical of sitting down with a DVD and an empty server, or even a really complicated, streamlined build-biased datacentre with PXE boot support and virtual disks and the rest, and you will begin to get Microsoft’s point about the persuasive nature of both the service (which it grudgingly admits has been around for a while) and the comparatively new user Interface.
Why do I say it’s too easy? Because all this activity is charged by the minute, is why. Charges accumulate not just by a single measure, but across the variety of ways that a server can be loaded – so you could have a machine with the smallest possible processor and memory specification, and then accrue costs on storage space and bandwidth connecting back to your storage in your local, hybrid cloud. Each activity will attract costs, and if you are a grey-haired IT type like me then the ease of provisioning – on the order of a few minutes for a complete Server 2012 VM, a little bit longer if you use the template for more layered installs like BizTalk or even SUSE Linux – could easily lead you into budget-denting territory within a few minutes.
Yes, I did say that you could have Linux VMs within Azure. On my most recent trip to Redmond, Microsoft took pains to point out that these are “user community provided templates” – that is, they are not in any way endorsing the suitability or supportability of these VMs just because they appear in a provisioning list – but be fair, was this the place you would expect to find not just other operating systems, but also ones provided by an enthusiastic user community, who very obviously have moved past the old allegiance-driven days and on to a far more utility-driven approach?
Those with lots of experience in cloud Infrastructure will be remarking already that a template-driven provisioning process sounds a lot like VDI, the second-generation thin client architecture which constructs and then destroys a complete single-user VM as a user authenticates and eventually, logs out. Azure VMs are not like that, VDI uses “Gold images” as the basis for each ephemeral VM, incorporating patches, updates and different LOBs (Builds for varied application mixes in differing Lines Of Business, hence the acronym) into each Gold Master.
Azure is overwhelmingly a server-orientated platform, and there are two important things to remember about servers in this context. Firstly, servers may reasonably be expected to stick to a lazy update cycle, and not at the convenience of the update issuer: system admins are in charge, and they may opt to run their spread of VMs on a sequence for receiving updates. This makes sense, one of the most immediate uses for a low-disk, cloud-resident Server VM is to put your software on it, run all the updates, and see how well the application fares. Not so good? Press “Delete” and start over.
The other reason for Azure server images not being built up and torn down is that servers tend to have stuff left on them. VDI machines make network connections and get their user profile data, and local documents from central corporate resources. The way an Azure VM functions, you would need to put in place a whole lot of other Azure resources before you could even consider an on-demand loading strategy like that. It’s not difficult – in particular the definition of a site-to-site VPN to relate the Azure instance to your LAN is a piece of cake – but quite rightly, MS doesn’t see the point in slowing down that quickfire provisioning with a lot of apparently unrelated folderol.
So, time to get started. You gallop through the nice clear slide-box style UI in your Azure trial and make yourself a straightforward Server 2012 VM. You can immediately see some advantages, even before you investigate the other charms down the left side of the Azure Portal pages.
Right from the off, your VM has a web reachable address – I decided to give my disposable VM an obscure word because I can see that the cloudapp.net TLD is going to suffer from dictionary erosion at speed as the service takes off.
Working out which ports to open, what updates to apply, and so on, is the familiar Windows Server hunk of your brain – though there are those charms on the left of the screen to make you wonder slightly. Why would I add an IIS role to my Azure VM, when I could just have a website space hosted on Azure proper? I know, there are plenty of good reasons to stick to known strategies and approaches when you have a quarter-century of code and business working practices expressed via the build-up of your web server: But against that impulse is Microsoft’s dripfeed pricing model.
You pay for your VM; that is fair. You pay for your disk space used; bytes cost money, as does just leaving a VM online but shut down. You also pay for time, CPU, disk etc incurred by MS usual drizzle of OS updates – effectively, whatever you do; you pay.
That’s fair enough, pay-as-you-go is the whole point of cloud services. What isn’t so good is the lack of clarity: there are pages and pages of tariff data to look through. But having looked through them, I couldn’t readily relate these to the amount of work I needed to do, or what Microsoft thought might be typical for an Azure user, or for someone tempted to move away from Amazon.
I entirely agree that cloud Infrastructure like this gets you off the hook of lengthy budgeting, procurement, provision and configuration processes, but it shouldn’t be left entirely up in the air what your likely cloud compute cost will be, as if you were entering into a poker game with Microsoft as both the dealer and the player.
I can foresee a lot of people who have been lagging behind looking at Server 2012 having their first experience of both IaaS, and the new server OS, by using Azure for the first time too. I worry, though, that when the first bill arrives and the trial runs out, there will be some long faces.
