Designing for Availability in the cloud (2013)

The multi-tenant (shared) nature of cloud computing platforms makes them notoriously prone to failure, as early adopters have learned the hard way.

But as we proved with the first edition of this study, published in December 2011, it is possible to design applications hosted on these services for high availability.

The major change on the Australian IaaS landscape in recent months was the arrival of Amazon Web Services in an Australian data centre. This eliminated one of the major barriers to adoption of cloud services for many Australian companies: the need to keep their data in-country.

But the arrival of Amazon – and imminent arrival of Rackspace – represent only little of what has changed. Below we’ve summarised the major changes:

1. Free trials

Offering a “free trial” is an ancient marketing technique, but one year ago it wasn’t an option among locally-based IaaS offerings. Today it is almost standard. Aside from Rackspace, all of the offerings in our matrix offer some sort of free trial, be it a monetary amount of services, a “free for 90 days” offer, Amazon’s free usage tier for 12 months or Google AppEngine’s permanent free threshold.

All parties figure that “try-before-you-buy” is one of the best ways to convert cloud sceptics into paying customers. If you’ve just spent 12 months developing an application on AWS or AppEngine, how likely is it you’ll to want to move away?

2. AWS matures

If you do desire the flexibility to move away to a new service, Amazon have made it relatively easy. In 2011 we lambasted the company for not offering VM export – it was the only availability feature we felt was missing. AWS now supports VM export, and their import capability has also been beefed up, making it easy to import a VM from your existing VMware vSphere, Citrix Xen or Microsoft Hyper-V deployment.

Amazon continues to be the gold standard that others judge themselves by.

3. Most improved: Microsoft Azure

We were impressed by how much Azure has improved. Microsoft now offers load balancing and DNS resolver capabilities, plus a form of VM import and export.

The Azure VM import involves uploading a virtual hard-disk created on an existing Hyper-V machine, which you can then use with your Azure virtual machine. There also appears to be a method of importing and exporting the VM configuration using PowerShell scripts, but import/export functionality isn’t clearly advertised as a feature of Azure.

The lack of server snapshots continues to be a big gap, but Azure is catching up to AWS on features and price. Azure is still slightly more expensive, but the gap is closing.

4. Rackspace re-invention

Rackspace have moved to OpenStack for their “next generation of Cloud Servers.” Their API isn’t backwards compatible with the first generation, so customers wanting to move existing services will need to recode. Not all the features of the first generation – such as scheduled backups – are available on the new platform.

In that sense, this study comes at an unfortunate time for Rackspace and we expect better results in future studies.

To their credit, Rackspace does offer some good documentation on reference architectures for a variety of use cases.

5. Stagnant in Sydney

Unfortunately, two of the three local offerings we studied in 2011 haven’t introduced any new availability features. Telstra’s offering is identical and OrionVM’s online feature list hasn’t been touched since May 2011.

The global pace of change in cloud services is rapid, and to see Australian companies standing still in the face of aggressive overseas competition does not inspire confidence.

But there is some good news.

Ninefold now have multiple Availability Zones, have expanded their API and remain the most feature-complete of the local offerings.

They now actively compare their services to Amazon EC2. It is refreshing to see somebody on the local scene taking the competitive threat from Amazon seriously.

6. Google App Engine

We didn’t feature Google App Engine in the 2011 study as it offers a very different way of using the cloud – something more akin to Platform-as-a-Service. Rather than merely having someone else run your servers, network and storage, App Engine takes away any thought of infrastructure at all. You consume compute and storage services, somewhere out there on the Internet. You just write applications, and forget about servers completely.

It’s therefore difficult to compare App Engine with the other cloud providers on this list, but readers demanded we keep track of the Google service, so we’ve done our utmost.

We have marked several things as Not Applicable. For example, with App Engine, your application responds to web requests, and App Engine figures out how to run your application based on the compute resources at its disposal. There is no failover API because it’s not under manual control; App Engine takes care of it for you.